I AE S   I n t e r n at ion al  Jou r n al   of   Ar t if icial   I n t e ll ig e n c e   ( I J - AI )   Vol.   14 ,   No.   4 Augus 2025 ,   pp.   3109 ~ 3120   I S S N:  2252 - 8938 ,   DO I 10 . 11591/i jai . v 14 .i 4 . pp 31 09 - 3120             3109     Jou r n al  h omepage ht tp: // ij ai . iaes c or e . c om   Fe d e r at e d   d e e p   l e ar n in g   in t r u si on   d e t e c t io n   syste m   on   sof t w ar e   d e f i n e d - n e t w or k   b ase d   in t e r n e t   of   t h in gs       Heba   Dhi r ar ,   Al i   H .   Ha m ad   D e pa r tm e nt   of   I nf or ma ti on   a nd   C omm uni c a ti on s   E ngi ne e r in g,   Al - K hw a r iz mi  C ol le ge  of  E ngi ne e r in g ,   U ni ve r s it y of  B a ghda d ,     B a ghda d,   I r a q       Ar t icle   I n f o     AB S T RA CT   A r ti c le   h is tor y :   R e c e ived   Oc 4,   2024   R e vis e d   J a 29,   2025   Ac c e pted   M a r   15,   2025       The   i n t ern e t   of   t h i n g s   (I o T )   a n d   s o ft w are - d ef i n e d   n e t w o r k s   (S D N )   p l ay   a   s i g n i f i ca n t   ro l e   in   en h an c i n g   effi ci e n cy   an d   p ro d u c t i v i t y .   H o w ev er,   t h ey   en co u n t er   p o s s i b l e   ri s k s .   A r t i f i ci a l   i n t e l l i g e n ce   (A I)   h as   recen t l y   b een   emp l o y e d   in   i n t ru s i o n   d e t ect i o n   s y s t ems   (ID S s ),   s erv i n g   as   an   i m p o r t an t   i n s t ru m en t   fo r   i mp r o v i n g   s ecu r i t y .   N e v ert h el e s s ,   t h e   n e ces s i t y   to   s t o re   d a t a   on   a   ce n t ra l i ze d   s er v er   p o s es   a   p o t en t i a l   t h rea t .   Fed erat ed   l earn i n g   (FL )   ad d re s s e s   t h i s   p r o b l em   by   t rai n i n g   mo d el s   l o ca l l y .   In   t h i s   w o r k ,   a   n et w o rk   i n t ru s i o n   d e t ect i o n   s y s t em   (N I D S)   is   i mp l eme n t e d   on   m u l t i - c o n t ro l l er   SD N - b as e d   Io T   n e t w o rk s .   The   i n t er p l a n et ar y   fi l e   s y s t em   ( IPFS)   FL   h as   b een   emp l o y e d   to   s h are   an d   t rai n   d ee p   l ear n i n g   (D L )   mo d el s .   Sev eral   cl i en t s   p art i ci p at e d   in   the   t rai n i n g   p r o ces s   u s i n g   c u s t o m   g e n era t ed   d at a s et   Io T - SD N   by   t ra i n i n g   t h e   mo d el   l o ca l l y   an d   s h ari n g   t h e   p arame t ers   in   an   e n cry p t e d   fo rmat ,   i m p ro v i n g   t h e   o v eral l   effect i v e n es s ,   s afe t y ,   an d   s ec u ri t y   of   t h e   n et w o r k .   The   mo d el   h as   s u cces s fu l l y   i d en t i f i ed   s ev er al   t y p es   of   at t ac k s ,   i n c l u d i n g   d i s t ri b u t ed   d e n i a l   of   s erv i ce   (D D o S),   d en i al   o s erv i ce  (D o S) ,   b o t n e t ,   b ru t e   f o rce,   ex p l o i t a t i o n ,   mal w are,   p ro b e,   w eb - b a s ed ,   s p o o f i n g ,   reco n ,   an d   ach i e v i n g   an   accu rac y   of   9 9 . 8 9 %   an d   a   l o s s   of   0 . 0 0 5 .     K e y w o r d s :   F e de r a ted   de e p   lea r ning   I nter ne t   of   thi ngs   I nter plane tar y   f il e   s ys tem   I ntr us ion   de tec ti on   s ys tem   S of twa r e - de f ined   ne twor k   Th i s   is   an   o p en   a c ces s   a r t i c l e   u n d e r   the   CC   BY - SA   l i ce n s e.     C or r e s pon din g   A u th or :   He ba   Dh i r a r   D e p a r t men t   of   I n f o r mat i on   a nd   C o mm un ica t io n s   E n g ine e r i ng Al - K hw a r iz m i   C o ll e g e   o f   E n g ine e r i ng   U n ive r s it y   o f   B a g hd a d   J a dr iaa ,   B a ghda d,   I r a q   E mail:   he ba . d@ke c bu. uoba ghda d. e du . iq       1.   I NT RODU C T I ON     T he   ter m   " int e r ne t   of   thi ng s   ( I o T ) "   r e f e r s   to   c onn e c ti ng   e mbedde d   de vice s   to   the   int e r ne t .   T he   idea   be hind   the   I oT   is   to   e na ble   e ve r yda y   i tems   to   be   c onne c ted   ove r   the   ne twor k   a nd   ga the r   va s t   a mount s   of   da ta   f r om   de vice s   with   dif f e r e nt   powe r s   a nd   li mi ted   r e s our c e s ;   he nc e ,   e nf or c ing   s e c ur it y   a nd   pr otec ti o n   can   be   c ha ll e nging   [ 1] .   Ne two r k   tr a f f ic   a na lys is   a nd   a bnor mal   a c ti vit y   identif ica ti on   a r e   r e s our c e - int e ns ive   tas ks .   Ove r   the   las t   s e ve r a l   ye a r s ,   s e ve r a l   li gh twe ight   a ppr oa c he s   f or   im pr oving   I o T   s e c ur it y   ha ve   be e n   c r e a ted     [ 2] ,   [ 3] ,   but   thes e   s ys tems   a r e   una ble   to   ha ndle   the   s igni f ica nt   s e c ur it y   r is ks   that   ha ve   be e n   dis c ove r e d   late ly.   He nc e ,   it   is   im pe r a ti ve   to   de s ign   e f f e c ti ve   in tr us i on   de tec ti on   to   e f f icie ntl y   de f e nd   a ga in st   va r ious   f or ms   of   a tt a c ks .   I ntr us ion   de tec ti on   s ys tem   ( I DS)   pe r f or m s   an   e s s e nti a l   pa r t   as   the   pr im a r y   de f e ns e   mec h a nis m   [ 4]   whic h   e mpl oys   many   a ppr oa c he s   to   identif y   a nd   f lag   a bnor malit ies .   S ign if ica nt   a dva nc e ments   ha ve   be e n   a c hieve d   us ing   mac hine   lea r ning   ( M L )   a nd   de e p   lea r ning   ( DL )   in   r e c e nt   ye a r s ,   r e s ult ing   in   wide s pr e a d   us e   a c r os s   s e ve r a l   domains .   It   can   of f e r   tec hniques   to   identif y   va r ious   f or ms   of   a tt a c k   without   the   ne e d   f or   s igni f ica nt   human   invol ve ment .   W hil e   thes e   methods   ha ve   pr ove n   e f f e c ti ve   f or   I DS ,   they   of ten   ne e d   a   c e ntr a li z e d   s e r ve r   to   a na lyze   the   da ta   ga ther e d   f r o m   a ll   ne twor k   us e r s .   F e de r a ted   lea r ning   ( F L )   is   a   mea ns   to   im pleme nt   on - de vice   lea r ning   while   pr e s e r ving   da t a   pr ivac y   [ 5 ] [ 7] .   FL   is   an   it e r a ti ve   p r oc e dur e   in   w hich   the   Evaluation Warning : The document was created with Spire.PDF for Python.
                                I S S N :   2252 - 8938   I nt  J   Ar ti f   I ntell Vol.   14,   No.   4,   Augus 2025 310 9 - 3120   3110   e nti r e   model   may   be   e nha nc e d   in   each   r ound   by   tr a ini ng   the   model   on   many   de vice s   a nd   us ing   th e ir   da ta   ac r os s   numer ous   it e r a ti ons   without   e xc ha nging   da ta   with   a   c e ntr a li z e d   s e r ve r   a c hieving   pr ivac y   pr e s e r va ti on   a nd   c os t   r e duc ti on,   as   e xpe c ted   in   c onve nti ona l   c e ntr a li z e d   a ppr oa c he s   [ 8] .   S of twa r e - de f ined   ne t wor king   ( S DN )   is   an   innovative   a r c hit e c tur e   that   s e pa r a tes   ne twor k   c ontr ol   f r om   f or wa r ding   f unc ti ons ,   e na bli ng   dir e c t   pr ogr a mm a bil it y   of   ne twor k   mana ge ment   a nd   e n ha nc ing   ope r a ti ona l   e f f icie nc y.   T he   S DN   ne twor k   uti li z e s   thes e   a tt r ibut e s   to   c r e a te   a   pr oa c ti ve   s ys tem   f or   de tec ti ng   int r us ions   in   I oT   ne twor ks ,   making   it   a   s upe r ior   c hoice   f or   ove r c omi ng   the   c ha ll e nge s   f a c e d   in   the   e f f icie nt   ope r a ti on   of   I oT   due   to   its   pr ogr a m mabili ty   a nd   c ompr e he ns ive   pe r s pe c ti ve   [ 9] ,   [ 10] .     T he   ne twor k - ba s e d   I DS,   r e f e r r e d   to   as   ne twor k   i ntr us ion   de tec ti on   s ys tem   ( NI DS ) ,   is   de s igned   to   de ter mi ne   whe ther   IP   tr a f f ic   is   c ompr o mi s e d   by   th r e a ds .   T he   p r oc e s s   c ons is ts   of   a   t r a ini ng   pha s e   uti li z ing   an   a c c ur a te   r e pr e s e ntation   of   r e c ognize d   a c ti vit ies ,   f o ll owe d   by   an   ope r a ti ona l   c las s if ica ti on   a nd   de c is ion   pha s e .   T he   tr a ini ng   a nd   c las s if ica ti on   pha s e s   r e ly   on   the   de f ini ti on   a nd   e xtr a c ti on   of   a   s e t   of   s tatis ti c a l   pa r a mete r s   a s s oc iate d   with   e a c h   IP   f low,   whic h   c ons ti tut e   th e   s tatis ti c a l   f inger pr int   of   the   f low ,   a nd   on   DL   c las s if ier s   de s igned   to   dif f e r e nti a te   be twe e n   nor mal   a nd   malicious   tr a f f ic.   In   thi s   s tudy,   FL   wa s   uti li z e d   to   c oope r a ti ve ly   tr a in   DL   models   to   im p leme nt   a nomaly - ba s e d   I DS   on   a   mul t i - c ontr oll e r   S DN - ba s e d   I oT   that   leve r a ge s   the   c ha r a c ter is ti c s   of   S DN   to   e s tablis h   a   p r oa c ti ve   s ys tem   f or   de tec ti ng   int r us ions   in   I o T   ne twor ks .   S e ve r a l   c li e nts   can   obtain   the   DL   model   f r om   the   in ter pl a ne tar y   f il e   s ys tem   ( I P F S )   ne twor k   a nd   pa r ti c ipat e   in   the   tr a ini ng   pr oc e s s   by   tr a ini ng   the   model   loca ll y   on   their   c us tom - ge ne r a ted   da tas e t   a nd   s ha r ing   only   the   pa r a mete r s   in   an   e nc r ypted   f or m   us ing   a dva nc e d   e nc r ypti on   s tanda r ds   ( AE S )   a lgor it hm.   T his   pr oc e s s   e nha nc e s   the   ove r a ll   e f f icie nc y,   s a f e ty,   a nd   s e c ur it y.   T he   model   ha s   s uc c e s s f ull y   identif ied   s e ve r a l   types   of   a tt a c ks   a c hieving   an   a c c ur a c y   of   99. 89 %   a nd   a   los s   of   0. 005 .     T he   r e s t   of   the   pa pe r   is   or ga nize d   a s   f oll ows :   s e c t ion   2   f oc us e s   on   I DS   r e s e a r c h   in   c ontext   of   S DN   a nd   I oT   ne twor ks ,   a nd   s e c ti on   3   c ontains   c omp r e he ns ive   ba c kgr ound   a na lys is   with   the   de tails   of   our   c us tom - ge ne r a ted   da tas e t   I oT - S DN .   T he   s ugge s ted   methodology   is   dis c us s e d   in   s e c ti on   4.   T he   e xpe r im e ntal   r e s ult s   a nd   e f f ica c y   of   pr opos e d   method   a r e   pr e s e nted   in   s e c ti on   5,   whe r e a s   s e c ti on   6   ou tl ines   the   wor k's   c on c lus ion.         2.   RE L AT E D   WORK   T he   s ubs tantial   a mount   of   da ta   a nd   the   diver s it y   of   de vice s   make   the   s e c ur it y   of   the   I oT   a   s igni f ica nt   pr oblem.   I DSs   ha ve   be e n   de ve loped   e mpl oying   v a r ious   methodologi e s   a nd   s tr a tegie s   to   s e c ur e   a n d   de f e nd   I oT   ne twor ks .   S e ve r a l   p r omi ne nt   int r us ion   de te c ti on   a lgor it hms   r e c e ntl y   de ve loped   to   a ddr e s s   s e c ur it y   c ha ll e nge s   in   S DN   a nd   I oT   ne twor ks   a r e   outl ined   in   T a ble   1   whic h   pr ovides   a   s umm a r y   of   the   r e s e a r c he r s   who   ha ve   c onc e ntr a ted   on   im pleme nti ng   I DS   on   th e   S DN   ne twor k.         T a ble   1.   S ur ve y   of   the   mos t   r e late d   wor k   of   I D S s   on   the   S DN   a nd   I o T   ne twor ks   R e f e r e n c e   Y e a r   N e t w o r k   D a t a s e t   T e c hn i qu e   A c c ur a c y   ( % )   T a n g   e t   a l [ 11]   2016   S D N   N S L - K D D   D e e p   ne u r a l   ne t w or ( D N N )   75. 7 5   A j a e i y a   e t   a l [ 12]   2017   S D N   C us t o m   R a ndo m   f o r e s t   ( R F )   85. 4   Ye   e t   al [ 1 3]   2018   S D N   C us t o m   S uppo r t   ve c t or   m a c hi n e   ( S V M )   95. 2 4   L a t a h   a nd  T oke r   [ 14]   2018   S D N   N S L - K D D   D e c i s i o t r e e   71   T a n g   e t   a l [ 15]   2019   S D N   N S L - K D D   G a t e   r e c ur r e n t   un i t   ( G R U ) -   r e c u r r e n t   ne u r a l   n e t w or k   ( R N N )   89   B o p p a n a   e t   al [ 16]   2019   S D N   N S L - K D D   RF   81. 9 5   H a n n a c h e   a n d   B a t o u c h e   [ 1 7 ]   2020   S D N   C us t o m   DNN   96. 1 3   L i m   e t   a l [ 18]   2020   S D N - I oT   N .   A .   FL - R F   w i t a c t o r - c r i t i c   P P O   N . A   E l S a y e d   e t   a l [ 19]   2021   S D N   I nS D N   [ 20]   C onvo l ut i on a l   ne ur a l   ne t w or ks   ( C N N ) + R F   99. 2 8   H a de m   e t   a l [ 21]   2021   S D N   N S L - K D D   S V M   95. 9 8   A l z a h r a ni   a n A l e n a z i   [ 22]   2021   S D N   N S L - K D D   X G B oos t   D e t e c t i on :   95 . 5 ,   C l a s s i f i c a t i on:   95 . 9 5   W a n i   e t   a l [ 23]   2021   S D N   C S E - C I C - I D S   20 18   I D S   I oT - S D L   99. 0 5   M oh s i n   a nd  H a m a d   [ 24]   2022   S D N   C us t o m   RF   K N N   N a i ve   B a ye s   ( N B )   L ogi s t i c   r e gr e s s i o ( L R )   R F :   100   K N N :   99 . 9 9 - 1 00   N B :   7 2. 1 1 - 8 3. 5   L R :   59. 44 - 92 . 74   R a vi   e t   al [ 25]   2022   S D N - I oT   S D N - I oT   [ 26]   G R U   f e a t ur e   f us i on   D e t e c t i on :   99   C l a s s i f i c a t i on:   98   J os e   a nd   J o s e   [ 2 7]   2023   I oT   C I C - I D S   2017   D N N ;   L S T M ;   C N N   94. 6 1;   97 . 67 ;   98 . 61   L oge s w a r i   e t   a l [ 28]   2023   S D N   N S L - K D D   H F S - L G B M   98. 7 2   C ha ga n t i   e t   a l [ 29]   2023   S D N - I oT   S D N   I oT - f oc u s e d   L S T M   97. 1   M a d du  a nd   R a o   [ 30]   2023   S D N   I nS D N   e dg e   I I oT   DL   99. 6 5   E l s a ye d   e t   al [ 3 1]   2023   S D N - I oT   T oN - I oT   I nS D N   L S T M   96. 3 5;   99 . 73   V i dhy a   a nd   N a g a r a j a [ 32]   2024   S D N - I oT   C S E - C I C - I D S 2018 ;   S D N - I oT   B i L S T M - b a s e W N I D S   99. 9 7 - 9 9. 96   95. 1 3 - 9 2. 90   N i kna m i   a nd   W [ 33]   2024   S D N   N S L - K D D ;   K D D 99   D e e pI D P S   ( C N N - L S T M + A M )   92. 2 - 95 . 4 ;   95 . 26 - 97 . 42   O ur   w o r k   2024   S D N - I oT   I oT - S D N   [ 34]   FL - DL   99. 8 9   Evaluation Warning : The document was created with Spire.PDF for Python.
I nt  J   Ar ti f   I ntell     I S S N:   2252 - 8938       F e de r ated  de e lear ning  int r us ion  de tec ti on  s y s te on  s oft w ar e   de fi ne d - ne tw or k   …  ( He ba  Dhir ar )   3111   Aja e iya   et   al.   [ 12]   int r oduc e d   RF   ba s e d   I DS   f or   i de nti f ying   ne twor k   thr e a ts   in   S DN .   T he   ne twor k   f e a tur e s   us e d   to   tr a in   the   model   a nd   pr e dict   ne twor k   a tt a c ks   c ons is ted   of   tupl e - 5,   pa c ke t   c ount,   byte   c ount,   a nd   pa c ke t   int e r a r r ival   ti me.   T he   de tec ti on   metho d   wa s   tes ted   a ga in s t   many   types   of   a tt a c ks ,   including   br ute   f or c ing,   por t   s c a nning,   a nd   f loodi ng   a tt a c ks .   W hil e   the   r e s ult s   indi c a ted   a   high   leve l   of   a c c ur a c y   in   de tec ti ng   a tt a c ks   us ing   the   RF   method,   ther e   is   a   lac k   of   de tailed   inf or mation   on   the   s e lec ti on   of   the   da tas e t   f or   a tt a c k   tr a f f ic.   T he r e f or e ,   thes e   r e s ult s   may   be   s olely   r e leva nt   to   non - I oT   tr a f f ic.   Ye   et   al .   [ 13]   int r oduc e d   a   dis tr ibut e d   de nial   of   s e r vice   ( DD oS )   a tt a c k   de tec ti on   s ys tem   in   S DN   that   uti li z e d   S VM .   T he   f e a tur e   s e t   us e d   f or   pr e dictin g   f loodi ng   a tt a c ks   c ons is ted   of   the   6 - tupl e   ne tw or k   f low   c ha r a c ter is ti c s .   T he   a uthor s   s tate   that   they   a c hi e ve d   an   a ve r a ge   de tec ti on   a c c ur a c y   r a te   of   95 . 24%   in   de tec ti ng   us e r   da tagr a m   pr otocol   ( UD P )   f loodi ng   a tt a c ks .   How e ve r ,   the   a tt a c k   tr a f f ic   c r e a ted   with   th e   hping3   tool   is   not   s uit a ble   f or   ge ne r a ti ng   I o T   tr a f f ic.   L a tah   a nd  T oke r   [ 14]   c onduc ted   a   c ompar is on   of   s e ve r a l   s upe r vis e d   ML   methods   f or   a nomaly - ba s e d   int r us ion   de tec ti on   in   S DN s .   T he   a uthor s   s tate d   that   the   de c is ion   tr e e     a lgor it hm   ob taine d   a   higher   a c c ur a c y   of   99. 7 %   whe n   the   ne twor k   s e c ur it y   labor a to r y   ( N S L ) - KDD   da tas e t   c ha r a c ter is ti c s   we r e   uti li z e d   as   i nput   f or   c ompar ing   ML   de tec ti on   models .   How e ve r ,   the   dis ti nc ti ve   c ha r a c ter is ti c s   of   S DN   f o r   de tec ti ng   a nomalies   s hould   be   take n   int o   a c c ount.   Ne ve r thele s s ,   the   NSL - KDD   da ta s e t   wa s   s pe c if ica ll y   c r e a ted   to   a s s e s s   a nd   identif y   tr a dit ional   ne twor k   tr a f f ic,   r a ther   than   f oc us ing   on   the   c a pa bil it ies   of   S DN .   B oppa na   et   al .   [ 16]   c onduc ted   a   c ompar is on   of   ML   a lgor it hms   us ing   va r ious   f e a tur e   s e lec ti on   methods   in   the   S DN   a nomaly   de tec ti on   modu le.   T he   NSL - KDD   da tas e t   wa s   uti li z e d   to   a s s e s s   the   e f f e c ti ve ne s s   of   va r ious   f e a tur e   a nd   ML   model   c ombi na ti ons   in   the   c ontext   of   S DN .   How e ve r ,   th e   a uthor s   a c knowle dge   that   c onduc ti ng   tes ts   on   a   r e a l - time   S DN   tes tbed   is   a   potential   f utu r e   goa l   to   ve r if y   th e   va li dit y   of   their   f indi ngs .   Ha de m   et   al .   [ 21]   ut il ize d   an   S VM   a nd   s e lec ti ve   loggi ng   with   IP   t r a c e ba c k   to   a c c ur a tely   identi f y   a tt a c ks   in   S DN   us ing   an   I DS   whic h   a ls o   he lped   c ons e r ve   memor y   r e s our c e s .   T he   NSL - KDD   da tas e t   uti li z e d   yielde d   a   de tec ti on   a c c ur a c y   of   87. 74% .   How e ve r ,   the   da tas e t   is   not   s our c e d   f r o m   non - I oT   ne two r ks ,   a nd   ther e   is   s ti ll   potential   f or   e nha nc ing   a c c ur a c y.   Alz a hr a ni   a nd  Ale na z i   [ 22 ]   pr e s e nted   a   NI DS   f o r   S DN s   that   us e s   the   e xtr e me   gr a dient   boos ti ng   ( XG B oos t)   model   to   a c c ur a tely   c a tegor ize   ne tw or k   int r us ions .   F ive   f e a tur e s   we r e   c hos e n   f r om   41   in   the     NSL - KDD   da tas e t.   T he   given   f ind ings   indi c a te   a   c las s if ica ti on   a c c ur a c y   of   95 . 5%   f o r   XG B oos t.   Additi ona ll y,   the   a uthor s   e mphas ize   that   their   a ppr oa c h   may   be   us e d   f or   S DN .     M ohs in  a nd  Ha mad  [ 24]   inves ti ga ted  the  e f f e c ti ve ne s s   of   va r ious   s upe r vis e ML   a lgor it hms   f or   de tec ti ng  DD oS   a tt a c ks   a c r os s   dif f e r e nt   S DN   ne twor topo logi e s .   T he y   a ppli e d   R F ,   k - ne a r e s n e ighbor s   ( KN N) ,   NB ,   a nd  L R   to   s ingl e ,   li ne a r ,   a nd   mul ti - c ontr oll e r   a r c hit e c tur e s .   T he i r   r e s ult s   s howe that  while  R F   a nd  KN a c hieve s tr ong  de tec ti on  pe r f or manc e ,   NB   a nd  LR   s uf f e r e f r om   low  a c c ur a c a nd   a   hig r a te  of   f a ls e   pr e dictions ,   li mi t ing  t he ir   s uit a bil it f or   pr a c t ica de ploym e nt.   J os e   a nd  J o s e   [ 27]   inves ti ga ted   the   e f f ica c y   of   DNN,   c onvolut ional   ne ur a l   ne twor ks   ( C NN ) ,   a n d   long   s hor t - ter m   memor y   ne twor ks   ( L S T M )   in   I o T   e nvi r onments   f or   the   de ploym e nt   of   I DS   ut il i z ing   the     C I C - I DS   2017   da tas e t.   T he   r e s ult s   indi c a ted   that   DL   models   ou tper f or med   pr e vious   methods   us e in   I o T - ba s e I DS.   S pe c if ica ll y,   L S T M   a nd  C NN   a c hieve a c c ur a c ies   of   97. 67%   a nd  98. 61% ,   r e s pe c ti ve ly,   while  the  ove r a ll   DL   a ppr oa c r e a c he 94. 61 %   a c c ur a c y.   T he   a f o r e mentioned   s tudi e s   togethe r   e it he r   im it a t e   the   be ha vior   of   c onve nti ona l   ne twor k   tr a f f ic   or   e mpl oy   the   a tt r ibut e s   of   pr e vious   ne twor k   tr a f f ic   da ta   to   pe r f or m   tes ti ng.   T he   e xpe r im e nts   c onf ir m   that   it   is   pos s ibl e   to   int e gr a te   s uc h   e nha nc e ments   int o   the   modul e   that   is   in   c ha r ge   of   de tec ti ng   a tt a c ks   in   t he   S DN   c ontr oll e r .   How e ve r ,   I o T   ne twor k   t r a f f ic   s hould   be   c ons ider e d,   as   it   is   pr oduc e d   thr ough   the   uti li z a ti on   of   I o T   de vice s   ins ide   the   S DN   f r a mew or k,   or   by   c ombi ning   the   f low   of   I o T   tr a f f ic   with   c onve nti ona l   ne twor k   tr a f f ic   to   e va luate   the   de tec ti on   e f f e c ti ve ne s s   of   ML   models .   F u r ther mor e ,   the   pe r f o r manc e   of   de tec ti ng   or   c las s if ying   in   s upe r vis e d   or   uns upe r vis e d   ML   models   s ti ll   ne e ds   e nha nc e ment   in   the   S DN   ne twor k.   Ne ve r thele s s ,   s ome   r e s e a r c he r s   inves ti ga t e d   f ur ther   the   uti li z a ti on   of   ne u r a l   ne twor k   models   f or   the   d e tec ti on   a nd   c a tegor iza ti on   of   ne two r k   a tt a c ks   in   S DN   s uc h   a s :     C ha ga nti   e al   [ 29]   a L S T M - ba s e a r c hit e c tur e   f or   int r us ion  de tec ti on  in   S DN - e na bled  I oT   ne twor ks .   T he ir   mod e e f f e c ti ve ly  identif ied  a nd  c las s if ied  va r ious   ne twor a tt a c ks ,   including  por s c a nning,   ope r a ti ng  s ys tem  f inger pr in ti ng,   de nial  of   s e r vi c e   ( DoS) ,   a nd  DD oS .   T he   r e s ult s   highl ight   the   model’ s   s uit a bil it f o r   c a ptu r ing  tempor a l   pa tt e r ns   a n e nha nc ing  de tec ti on   a c c ur a c in   c ompl e x   S DN - I oT   e nvir onments .   E ls a ye d   et   al .   [ 31 ]   c onduc ted   a   s e c ur e d   a utom a ti c   two - leve l   int r us ion   de tec ti on   s ys tem   ( S AT I DS)   that   e mpl oye d   an   e nha nc e d   L S T M   ne twor k   a nd   uti li z e d   T oN - I oT   a nd   I nS DN   da tas e ts .   T he   a uthor   s tate d   that   the   pr opos e d   s ys tem   e f f e c ti ve ly   dis ti nguis he d   be twe e n   malicious   a nd   ha r ml e s s   ne twor k   tr a f f ic,   a c c ur a tely   c a tegor ize d   the   type   of   a tt a c k,   a nd   pr e c is e ly   identif ied   the   s pe c if ic   s ub - a tt a c k.   T he   r e s e a r c h   r e s ult s   de mons tr a ted   that   the   s ugge s ted   s ys tem   s ur pa s s e s   other s   in   identi f ying   a   wide   r a nge   of   a tt a c ks .   How e ve r ,   L S T M - ba s e d   models   ne e d   s ubs tantial   memor y   c a pa c it y   thr oughout   the   tr a ini ng   pr oc e s s .   T he   s u bs tantial   Evaluation Warning : The document was created with Spire.PDF for Python.
                                I S S N :   2252 - 8938   I nt  J   Ar ti f   I ntell Vol.   14,   No.   4,   Augus 2025 310 9 - 3120   3112   memor y   r e s our c e   c ons umpt ion   mi ght   r e s tr ict   the   uti li z a ti on   of   L S T M   f o r   I DS   in   S DN   a nd   I o T   n e twor ks .   Als o,   in   a   c ompl e x   I o T   ne twor k ,   the   s ugge s ted   a r c hit e c tur e   r e quir e s   s igni f ica nt   ti me   to   tr a in   the   mod e l   due   to   the   pr oc e s s   of   s e lf - lea r ning   the   f e a tur e s   a nd   a djus ti ng   the   model   we ight s .       3.   T E CHNOL OG Y   B AC KG ROUN D   B e f or e   e xa mi ning   the   p r opos e d   model,   it   is   e s s e nti a l   to   ge t   an   unde r s tanding   of   the   main   tec hnique   a nd   method   uti l ize d   in   thi s   s tudy.   W hich   wa s   s e lec ted   thr ough   an   e va luation   of   the   p r ior   s tudi e s   that   c ons ider   the   de ve lopm e nt   of   an   e f f icie nt   I DS   s ys tem   a nd   a na lyze   the   us e d   too ls .   T his   s e c ti on   pr ovides   an   ov e r view   of   the   tec hnologi e s   a nd   methodologi e s   uti li z e d   to   im p leme nt   NI DS   on   an   S DN   ne twor k   as   f o ll ows .     3 . 1.     S o f t war e   d e f i n e d   n e t wor k     S witche s   a nd   r outer s   we r e   uti li z e d   in   tr a dit ional   ne twor ks   to   e s tablis h   ne twor k   c onne c ti ons   a nd   f a c il it a te   the   tr a ns mi s s ion   of   da ta   th r oughout   the   n e twor k.   T his   ne twor king   tec hnique   may   be   vulne r a ble   to   a   lac k   of   c onf identialit y   a nd   s us c e pti ble   to   thi r d - pa r ty   a tt a c ks .   S DN   is   a   ne twor king   s tr a tegy   that   e nha nc e s   the   e f f icie nc y   of   a   c e nt r a li z e d   e nvi r onment   by   s e p a r a ti ng   da ta   tr a ns f e r   f r om   de dica ted   de vice s   [ 3 5] .   T his   pa r a digm   is   s tr uc tur e d   a r ound   dis ti nc t   plane s ,   each   with   its   own   de s ignate d   f unc ti ons i )   d a t a   plane   r e s pons ibl e   f or   the   f or wa r ding   of   pa c ke ts ;   ii )   t he   c ontr ol   plane   de ter mi ne s   r outi ng   by   leve r a ging   a   f l ow   table   that   pr ovides   r ules   f or   e f f icie ntl y   mana ging   inco mi ng   pa c ke ts ;   a nd  T he   a ppli c a ti on   plan   c ontains   a   r a nge   of   s e r vice s   that   a r e   of f e r e d   to   us e r s .     How e ve r ,   ne w   vulner a bil it ies   may   a ls o   be   int r oduc e d   f r om   th is   s e pa r a ti on.   F o r   e xa mpl e ,   the   c ontr oll e r   can   be   il lus tr a ted   by   e xha us ti ng   the   c o mm unica ti on   ba ndwidth   be twe e n   inf r a s tr uc tur e   la ye r s   s uc h   as   the   Ope nF low   s witch   a nd   S DN   c ontr oll e r .   Ne ve r thele s s ,   S DN   can   im pr ove   ne twor k   s e c ur it y   d ue   to   its   pr ogr a mm ing   c a pa bil it ies   that   e na ble   the   c r e a ti on   of   s e c ur it y   a ppli c a ti ons   s uc h   a s   I DS   that   de tec t   ne twor k   thr e a ts .   Als o,   it   is   im por tant   to   mention   that   f lo w   r ules   may   be   modi f ied   ba s e d   on   r e quir e ments   [ 36]   by   leve r a ging   the   a bil it y   to   pr og r a m   a nd   c ontr ol   o f f e r e d   by   S DN   in   c ompar is on   to   tr a dit ional   ne t wor king   s ys tems   [ 37] .   F igur e   1   i ll us tr a tes   the   typi c a l   S DN   a r c hit e c tur e .             F igur e   1.   S DN   ne two r k   a r c hit e c tur e       3 . 2.     I n t r u s ion   d e t e c t ion   s ys t e m s     I DS   is   a   c r uc ial   e leme nt   in   s a f e gua r ding   s ys tem s   by   de tec ti ng   a nd   a na lyzing   ne twor k   tr a f f ic   to   identif y   s e c ur it y   b r e a c he s   a nd   thr e a ts   us ing   one   of   the   f oll owing   tec hniques :   s ignatur e - ba s e d   or   a nomaly - ba s e d.   T he   f ir s t   method   r e li e s   on   pr e de ter mi ne d   ne twor k   pa tt e r ns   a nd   is   ther e f or e   una ble   to   identif y   ne w   a tt a c ks .   In   c ontr a s t,   the   latte r   method   a na lyze s   pa r ti c ular   c ha r a c ter is ti c s   of   ne twor k   t r a f f ic,   a ll o wing   a ny   diver ge nc e   f r om   nor mal   ne two r k   a c ti vit y   to   be   r e c ognize d   as   a   potential   a tt a c k;   a   s im ple   c ompar is on   be twe e n   them   is   pr e s e nted   in   T a ble   2,   [ 38] .   Ne ve r thele s s ,   s ome   dr a wba c ks   we r e   a ls o   int r oduc e d,   s uc h   as   th e   lac k   of   identif ica ti on   of   e nc r ypted   pa c ke ts   a nd ,   the   incide nc e   of   f a ls e   a lar ms   may   be   e leva ted,   lea ding   to   the   ne e d   f or   human   int e r ve nti on   to   a djus t   the   a nomaly   indi c a tor s   a nd   ult im a tely   r e s ult ing   in   an   inef f icie nt   s e c ur it y   s olut ion   [ 39] .     Evaluation Warning : The document was created with Spire.PDF for Python.
I nt  J   Ar ti f   I ntell     I S S N:   2252 - 8938       F e de r ated  de e lear ning  int r us ion  de tec ti on  s y s te on  s oft w ar e   de fi ne d - ne tw or k   …  ( He ba  Dhir ar )   3113   T a ble   2.   De tec ti on   tec hnique   c ompar is on   F a c to r   D e te c ti on   te c hni que   S ig na tu r e   A noma ly   A la r m   r a te   L ow   H ig h   S pe e d   H ig ht   L ow   F le xi bi li ty   L ow   H ig h   R e li a bi li ty   H ig h   M ode r a te   S c a la bi li ty   L ow   H ig h   R obus tn e s s   L ow   H ig h       R e c e ntl y,   s e ve r a l   ML   a ppr oa c he s   ha ve   be e n   int r oduc e d   to   identif y   int r us ion   in   S DN   a nd   I oT   [ 15 ] ,   [ 16] ,   [ 19 ] - [ 21] .   Als o,   the   DL   model   wa s   p r opos e d   in   the   c ontext   of   S DN   a nd   the   I o T   to   e nha nc e   int r us ion   a tt a c k   de tec ti on   [ 17] ,   [ 22] ,   [ 25 ] .   P r e vious   r e s e a r c h   on   int r us ion   de tec ti on   ha s   de mons tr a ted   that   the   DL   model   pr ovides   s upe r ior   pe r f or manc e   whe n   a ppli e d   to   lar ge - s c a le   ne twor k   da ta s e ts   [ 22] ,   [ 25] .   De s pit e   the   s ubs tantial   im pa c t   of   ML   a nd   DL   on   pr a c ti c a l   pr oblem - s olvi ng,   they   a r e   s ubjec t   to   many   li mi tations ,   including:   i )   u s e r s   mus t   pr ov ide   thei r   da ta   to   a   c e ntr a li z e d   s e r ve r   to   tr a in   the   model;   ii )   w he n   ne twor k   s ize   incr e a s e s ,   the   pe r f or manc e   dim ini s he s   a nd   ther e   is   a   r is k   of   a   s ingl e   poin t   of   f a il u r e   that   m ight   unde r mi ne   the   int e gr it y   a nd   qua li ty   of   s e r vice s   ( QoS ) ;   ii i)   I DS   ne e ds   r a pid   a na lys is ,   howe ve r ,   c e ntr a li z e d   pr oc e s s ing   is   a   time - c ons umi ng   pr oc e s s ;   a nd  iv)   I oT   de vice s   f r e que ntl y   ga ther   da ta   f r om   e nd - us e r s ,   potentially   e xpos ing   their   s e ns it ive   inf or mation.   To   tac kle   thes e   pr oblems ,   it ’s   ne c e s s a r y   to   us e   method s   that   invol ve   on - de vice   lea r ning.     3 . 3.     F e d e r at e d   lear n in g     Google   int r oduc e d   the   c onc e pt   of   FL   to   p r e s e r ve   da ta   pr ivac y   on   de vice s   [ 5] [ 7]   by   a ll owing   node s   to   lea r n   c oll a bo r a ti ve ly   without   s ha r ing   da ta   with   a   c e ntr a li z e d   s e r ve r .   FL   is   an   it e r a ti ve   p r oc e dur e   in   whic h   the   e nti r e   model   is   e nha nc e d   in   each   r ound   unti l   a   s pe c if ic   number   ha s   be e n   r e a c he d   or   the   r e quir e d   leve l   of   pe r f or manc e   is   a tt a ined.   In   the   be ginni ng ,   the   FL   s e r ve r   s e lec ts   a   dis ti nc t   gr oup   of   c li e nts   to   pa r ti c ipate   in   the   tr a ini ng   pr oc e s s   a nd   dis tr ibut e s   its   global   model   to   them   [ 7] .   Onc e   the   global   model   is   obtaine d,   e a c h   c li e nt   e mpl oys   its   da ta   f o r   loca l   tr a ini ng   a nd   tr a ns mi ts   th e ir   a c quir e d   pa r a mete r s   ba c k   to   the   s e r ve r ,   as   il lus tr a ted   in   Fi gur e   2.   It   of f e r s   a   pr ivac y   pr otec ti on   tec hnique   that   e f f icie ntl y   uti li z e s   the   pr oc e s s ing   r e s our c e s   of   the   pa r it y   de vice   f or   model   t r a ini ng,   ther e by   pr e ve nti ng   t he   lea ka ge   of   pr ivate   in f or mation   dur ing   da ta   tr a ns f e r .   C ons ider ing   the   e nor mous   number   of   de vice s ,   the r e   a r e   a   lar ge   number   of   r e leva nt   da tas e t   r e s our c e s   that   can   be   e f f e c ti ve ly   uti li z e d .           F igur e   2.   F L   ove r view       Ge ne r a ll y,   FL   may   be   c a tegor ize d   int o   th r e e   types   ba s e d   on   the   dis tr ibut ion   of   c li e nts '   da ta:   ve r ti c a l,   hor izonta l,   a nd   t r a ns f e r   FL .   He r ts   f o r   lea r ning   ( H F L )   is   an   FL   tec hnique   in   whic h   the   da tas e ts   on   t he   c li e nts   s ha r e   the   s a me   f e a tur e   but   ha ve   s e pa r a te   obs e r va ti ons .   Ve r ti c a f e de r a ted  lea r ning  ( VFL ) ,   o f ten   r e f e r r e d   to   as   f e a tur e s - ba s e d   F L ,   in   whic h   da ta   f r om   s e ve r a l   do mains   is   uti li z e d   to   tr a in   a   global   model.   In   thi s   c ontext,   the   c li e nt   da tas e t   may   c ontain   identica l   obs e r va ti ons   b ut   with   va r ying   c ha r a c ter is ti c s .   As ide   f r om   HFL   a nd   VFL ,   ther e   is   a ls o   the   f e de r a ted  tr a ns f e r   lea r ning  ( F T L )   a r c hit e c tur e   pr e s e nted   in   [ 40] ,   whic h   is   a ppli c a ble   whe n   the   da tas e ts   on   the   de vice s   dif f e r   not   only   in   oc c ur r e n c e s   but   a ls o   in   c ha r a c ter is ti c s .   How e ve r ,   to   e ns ur e   pr ivac y,   s ome   pr oblems   mus t   be   a ddr e s s e d   in   the   im ple menta ti on   of   F L :   i )   it   is   im pe r a ti ve   to   gua r a ntee   that   the   Evaluation Warning : The document was created with Spire.PDF for Python.
                                I S S N :   2252 - 8938   I nt  J   Ar ti f   I ntell Vol.   14,   No.   4,   Augus 2025 310 9 - 3120   3114   tr a ini ng   model   us e d   doe s   not   dis c los e   us e r s '   c onf i de nti a l   inf or mation ;   ii )   s ince   the   tr a ini ng   pr oc e s s   pr oc e e ds   loca ll y   at   each   e nti ty   us ing   its   da tas e t.   T he r e f o r e ,   it   is   c r uc ial   to   gua r a ntee   that   only   a ll owe d   e nti ti e s   pa r ti c ipate   in   the   t r a ini ng   p r oc e s s   a nd   the   r e c e ived   model   upda tes   ha ve   be e n   tr a ns mi tt e d   by   th e m;   ii i)   tr a dit ional   ML   models   r e quir e   a   s ubs tantial   a moun t   of   da ta   to   a c hieve   outs tanding   pe r f or manc e .   How e ve r ,   in   a   dis pe r s e d   c ontext,   the   a c c e s s ibl e   da ta   on   each   de vice   is   mi nim a l .   C onve r s e ly,   c ons oli da t ing   a ll   da ta   in   a   c e ntr a li z e d   wa y   mi ght   lea d   to   s igni f ica nt   c os ts ;   a nd  iv)   the   da ta   s tor e d   on   s uc h   de vice s   may   no t   e xhibi t   dis ti nc t   a nd   s ymm e tr ica l   dis tr ibut ion   ( non - I I D)   c ha r a c ter is ti c s ;   tr a ini ng   thes e   da ta   s e ts   pos e s   a   s u bs tantial   c ha ll e nge .     3 . 4.     I oT - S DN   d a t as e t   T he r e   is   a   lac k   of   publi c ly   a c c e s s ibl e   da tas e ts   that   a r e   e xpli c it ly   de s igned   f o r   int r us ion   de tec ti on   in   S DN - ba s e d   I oT .   F or   thi s   wo r k,   a   c us tom - ge ne r a ted   da tas e t   wa s   uti li z e d.   T he   da tas e t   c omp r is e s   e ight y - s ix   a tt r ibut e s   withi n   a   s ize   of   ( 2. 7   GB )   c oll e c ted   f r om   s im ulate d   S DN - ba s e d   I oT   ne twor ks   withi n   t wo   f low   pr of il e s :   nor mal   a nd   a tt a c k   tr a f f ic   s uc h   as   botnet,   br ute   f or c e ,   DoS ,   DD oS ,   e xploi tation ,   malwa r e ,   M I R AI ,   pr obe ,   R 2L ,   UR 2,   we b - ba s e d,   s poof ing,   a nd   r e c on,   e mpl oye d   us ing   M e tas ploi t.   T a ble   3   pr e s e nts   the   c oll e c ted   tr a f f ic   c a tegor ies   togethe r   with   their   c or r e s ponding   r e c or d   number s .   T he   ne twor k   topol ogy   is   im pleme nted   us ing   M ini ne t   Wi F i   on   the   Ubuntu   20. 04   L T S   ope r a ti ng   s ys tem   c ons e nt   of   two   R yu   c ontr oll e r s   who   we r e   r e s pons ibl e   f or   mana ging   the   ope r a ti on   of   the   f o ur   Ope nF low   s witche s   that   c onne c ted   to   f our   s ubdomains .   E a c h   s ubdomain   c ompr is e s   a   pa i r   of   hos ts   a   s ingl e   a c c e s s   point ,   a nd   thr e e   wi r e les s   s tations .   T he   f ir s t   two   s ubdomains   e nc ompas s   a   va r iety   of   s e r vice s ,   s u c h   as   HT T P   a nd   F T P   s e r ve r s .   In   c ontr a s t,   the   las t   two   c ompr is e   many   wi r e les s   s e ns or   de vice s .   T he   ne twor k   t r a f f ic   is   c a ptur e d   us ing   W i r e s ha r k   a nd   c las s if ied   a c c or ding   to   its   f e a tur e s   e xtr a c ted   us ing   C I C F low M e ter .       T a ble   3.   Da ta   r e c or ds   number   f o r   each   tr a f f ic   g r ou p   G r oup   T r a f f ic   t ype   R e c or ds   N or ma l   H T T P S ,   H T T P ,   F T P ,   D N S ,   ma il ,   br ow s in g,   a nd  Y ou T ube   367,396   A tt a c k   D oS ,   D D oS ,   R 2L ,   B r ut e - F or c e ,   E xpl oi ta ti on,   Web - B a s e d,   B ot ne t   P r obe ,   R e c on,   S poof in g,   M a lwa r e   5,878,336   ( 367,396   f or   e a c h)       4.   P ROP OS E D   M E T HO DOL OG Y   T he   tec hnique   e mpl oys   a   s ys t e matic   a ppr oa c h   tha t   s tar ts   with   the   pr e c is e   de f ini ti on   of   the   r e s e a r c h   is s ue .   T he   c ombi na ti on   of   FL   with   DL   tec hniques   f or   a nomaly   incur s ion   de tec ti on   in   S DN - ba s e d   I oT   ne twor ks   is   e mer ging   as   a   potentially   unique   a p pr oa c h.   T his   s e c ti on   de li ne a tes   the   pr ojec ted   a r c hit e c tur e   il lus tr a ted   in   F igu r e   3   whic h   ha s   be e n   e xe c uted   in   t wo   pr incipa l   pha s e s   as   f oll ows           F igur e   3.   S DN - ba s e I oT   p r opos e s ys tem  a r c hit e c tur e   Evaluation Warning : The document was created with Spire.PDF for Python.
I nt  J   Ar ti f   I ntell     I S S N:   2252 - 8938       F e de r ated  de e lear ning  int r us ion  de tec ti on  s y s te on  s oft w ar e   de fi ne d - ne tw or k   …  ( He ba  Dhir ar )   3115   4 . 1.     De p loyi n g   an   S DN   in f r as t r u c t u r e   f or   a n   I o T   n e t wor k   One   of   the   s igni f ica nt  vulner a bil i ti e s   is   the  S D ne twor whe the   c ontr oll e r   is   e xploi ted   by   ove r whe lm ing  the   c omm unic a ti on   c a pa c it y   with   e xc e s s iv e   a nd   unde s ir e d   tr a f f ic,   lea ding   to   a   DD o S   a tt a c k.   How e ve r ,   ne twor k   s e c ur it y   can   be   e nha nc e d   by   its   pr ogr a mm ing   c a pa bil it ies ,   whic h   a ll ow   f or   the   de ve lopm e nt   of   s e c ur it y   a ppli c a ti ons   s uc h   as   I DS   that   can   identi f y   ne twor k   th r e a ts .   Our   s ugge s ted   a r c hit e c tur e   c ompr is e s   numer ous   objec ti ve s ,   whi c h   a r e   i)   a   mul ti - c ontr oll e r   S DN   ne twor k   wa s   e s tablis he d   uti li z ing   Z ooke e pe r   a nd   R e dis .   Z ooKe e pe r   will   p r ompt ly   o r ga nize   a nd   c oo r dinate   the   c ha nge   of   c ontr oll e r   r ole,   whe r e a s   a   ba c kup   c opy   f r om   the   f low   table   w il l   be   s tor e d   in   R e dis   s tor a ge .   T his   s e tup   s e r ve s   as   a   r obus t   f r a mew or k   to   p r e ve nt   ne two r k   f a il ur e .   If   the   mas ter   c ontr o ll e r   be c omes   inac ti ve ,   the   other   c ontr oll e r   r e tr ieve s   the   f low   e ntr ies   f r om   R e dis   s tor a ge   a nd   s moot hly   c onti nue s   ne twor k   ope r a ti ons ,   F igu r e   4   de mons tr a tes   the   mul ti - c ontr oll e r   im pleme ntation   s teps ii )   ing r e s s   a nd   e gr e s s   poli c ies   we r e   e mpl oye d   to   mana ge   a n d   c ontr ol   ne twor k   tr a f f ic;   i ii )   a ll   pa c ke ts   r e c e ived   by   the   c ontr oll e r   will   be   ini t ially   s e nt   to   the   I DS   s e r ve r   to   pr e dict   whe ther   the   tr a f f ic   r e c e ived   is   an   a tt a c k   or   nor mal   tr a f f ic.   How e ve r ,   by   f lood ing   the   c ontr oll e r ,   the   I D S   s e r ve r   will   a ls o   be   f looded.   To   pr e ve nt   thi s ,   DoS   a nd   D DoS   a tt a c ks   a r e   mi ti ga ted   onc e   a   thr e s hold   is   r e a c he d;   a nd  iv)   the   s lave   c ontr oll e r   is   uti li z e d   to   e f f icie ntl y   mana ge   the   huge   a mount   of   da ta   r e c e ived   on   th e   mas ter   c ontr oll e r   by   e na bli ng   P us hba c k   pol ice .   M ini ne t   W iF i   us e d   to   c ons tr uc t   a   tr e e   topol ogy   il lus tr a ted   in   F igu r e   4   c ons is ts   of   f our   domains .   E a c h   domain   is   c ompos e d   of   two   hos ts ,   an   a c c e s s   point ,   thr e e   s tations ,   a nd   thr e e   wir e les s   s e n s or s .   I oT   de vice s   may  e xpe r ienc e   c omm unica ti on  r e s our c e   li mi ts   that  pr e ve nt  the f r om   int e r a c ti ng  with  a   c e ntr a ba s e   s tation  due   to   the  li mi tations   in  c omm unica ti on  r e s our c e s .           F igur e   4.   Z ooKe e pe r   a nd   R e dis   c oor dination   s ys tem       4 . 2.     De p loyi n g   an om aly - b as e d   n e t wor k   in t r u s io n   d e t e c t ion   s ys t e m   I nit ially ,   the   c oor dinator   s e r ve r   uploads   the   global   model   pr e s e nted   in   T a ble   4,   to   the   IP F S   ne twor k   whic h   is   uti li z e d   to   e nha nc e   s e c ur e   model   a ggr e ga ti on,   e ns ur ing   that   only   a uthor ize d   c li e nts   invol v e d   in   the   tr a ini ng   pr oc e s s   can   a c c e s s   a nd   downloa d   the   gl oba l   model   ba s e d   on   a   s pe c if ic   ha s h   identi f ier .   I P F S   is   a   de c e ntr a li z e d   f r a mew or k   invol ving   pr o tocols ,   pa c ka ge s ,   a nd   c ompos a ble   s pe c if ica ll y   de s igned   to   ha ndle,   dir e c t,   a nd   tr a ns mi t   c ontent - a ddr e s s e d   da ta.   T he   s ys tem   is   both   r e s our c e - e f f icie nt   a nd   r e li a bly   c onve r ge s   to   c e ntr a li z e d   FL   f r a mew or ks   with   a   dr op   of   les s   than   1%   [ 41] .   T he   model   c ons is ts   of   f our   de ns e   laye r s   with   256,   128 ,   64 ,   a nd   32   ne u r ons ,   r e s pe c ti ve ly   in   a ddi ti on   to   the   input   a nd   ou tput   laye r s .   E a c h   laye r   is   f oll owe d   by   a   ba tch  nor maliza ti on  laye r   a nd   a   dr opou t   r a te   of   0. 5 .   T he   r e a s on   f or   a dding   thes e   laye r s   is   to   im pr ove   the   pe r f or manc e   a nd   ge ne r a li z a ti on   of   the   ne twor k.   T he   ba tch   nor maliza ti on   s tabili z e s   the   lea r ning   p r oc e s s   by   nor malizing   the   a c ti va ti ons   of   the   p r e c e ding   laye r .   T he   dr opout   laye r   is   uti li z e d   to   mi ti ga te   ove r f it ti n g   in   the   model   by   e nha nc ing   its   a bil it y   to   ge ne r a li z e   to   n e w   da ta   a nd   incr e a s ing   its   ove r a ll   r e s il ienc e .   T he   r e c ti f ied  li ne a r   unit   ( R e L U )   a c ti va ti on   f unc ti on   wa s   us e d   in   a ll   the   De ns e   laye r s   due   to   its   s im pli c it y,   e f f icie nc y,   a nd   a bil it y   to   a ddr e s s   the   va nis hing   gr a dient   pr oblem .   S of tM a x   a c ti va ti on   f unc ti on   wa s   us e d   in   the   out put   laye r   f or   mul ti - c las s   c las s if ica ti on   tas ks   to   ge ne r a te   a   pr oba bil it y   dis tr ibut ion   a c r os s   dif f e r e nt   c las s e s .         T a ble   4.   DL   tr a ini ng   model   A lg or it hm   L a ye r s   N e ur on   DNN   4   D e ns e ,   in   a ddi ti on   to   in put   a nd   out put   la ye r   256,   128,   64,   32     4   B a tc h   N or ma li z a ti on   la ye r       4   D r opout   la ye r     A c ti va ti on   f unc ti on   Re L U ,   S of tM a x     L os s   f unc ti on   C a te gor ic a l   c r os s - e nt r opy     O pt im iz e r   A da m     B a tc h - s iz e   256         T a ble   il lus tr a tes   the  c las s if ica ti on  r e por of   de tec ti ng  e a c a tt a c type  a f ter   tr a ini ng  the  model.   All   of   the   metr ics   de mons tr a te  a   s upe r ior   de gr e e   o f   e f f e c ti ve ne s s   in  a ll   types   of   t r a f f ic,   with   a   notable   e xc e pti on  Evaluation Warning : The document was created with Spire.PDF for Python.
                                I S S N :   2252 - 8938   I nt  J   Ar ti f   I ntell Vol.   14,   No.   4,   Augus 2025 310 9 - 3120   3116   be ing  the  us e r   to  r oot   ( U2R )   c a tegor y   whic h   c a be   e xplaine by  the  f a c that  other   c a tegor ies   f r e que ntl dis play  mor e   dis s im il a r it in   c ompar is on  to  no r ma tr a f f ic  pa tt e r ns .   I c ont r a s t,   the  U2R   a tt a c king  c l a s s   ha s   a   notable   s im il a r it to   the  s tanda r da ta   tr a f f ic.     T he   I D S   model  is   de ployed  on   a   de dica ted  s e r ve r   to   pe r f or m   int r us ion  de tec ti on   to   the  c ompl e te   ne twor k,   whic o f f e r s   s ubs tantial  a dva ntage s   in  ter ms   of   pe r f or manc e ,   s c a labili ty,   s e c ur it y,   a nd  mai ntena nc e .   I gua r a ntee s   that  the   c ontr oll e r   c a c onc e ntr a te  o it s   pr im a ry   ope r a ti ons ,   while   the   I DS   s e r ve r   is   f i ne - tuned   a nd   e xpa nde d   e xpr e s s ly   f or   e f f icie nt   int r us ion   d e tec ti on.   In   an   S DN ,   the   pr oc e s s   of   pa c ke t   f or w a r ding   is   ha ndled   dif f e r e ntl y,   whe n   a   hos t   s e nds   a   r e que s t   to   a nother   hos t,   it   is   f ir s t   f o r wa r de d   to   the   Ope vS witch  ( OVS s witch.   T he   s witch   c he c ks   if   ther e   is   a ny   in s tr uc ti on   to   pr oc e e d   with.   If   not,   the   pa c ke t   is   f o r w a r de d   to   the   c ontr oll e r   to   identi f y   the   opti mal   pa th .   In   ou r   wor k,   the   mas ter   c ontr oll e r   s e nds   the   pa c ke t   to   the   NI DS   s e r ve r ,   whic h   c ontains   the   tr a ined   model.   T his   m ode l   pr e dicts   whe ther   the   pa c ke t   is   nor mal   or   int r us ion.   In   the   c a s e   of   nor mal   tr a f f ic,   the   c ontr oll e r   identif ies   the   opti mal   pa th   to   f o r wa r d   it   to   the   de s ti na ti on,   s e nds   the   ins tr uc ti on   ba c k   to   the   s witch,   a nd   a dds   a   ne w   f l ow   e ntr y   in   R e dis   s tor a ge .   If   it   is   an   int r us ion   t r a f f ic   the   c ontr oll e r   a dds   a   f low   e ntr y   to   block   the   s our c e   ho s t.       T a ble   5.   DNN   tr a ini ng   model   T r a f f ic   T ype   P r e c is io n   R e c a ll   F1 - s c or e   B ot ne t   0.9984   0.9936   0.9960   B r ut e - F or c e   0.9993   0.9815   0.9903   D D oS - I C M P   1.0   1.0   1.0   D D oS - UDP   0.9991   0.9994   0.9992   D oS - S Y N   1.0   1.0   1.0   D oS - UDP   1.0   0.9998   0.9999   E xpl oi ta ti on   0.9936   0.9975   0.9956   M a lwa r e   0.9987   0.9944   0.9966   M ir a i   0.9999   0.9989   0.9994   N or ma l   0.9994   0.9975   0.9985   P r obe   1.0   0.9991   0.9995   R 2L - I M A P   1.0   1.0   1.0   R e c on - P in gS w e e p   0.9936   0.9984   0.9960   R e c on - S ni f f in g   0.9983   1.0   0.9991   S poof in g   1.0   1.0   1.0   U 2R   0.9893   1.0   0.9946   Web - A tt a c k   0.9908   1.0   0.9953       5.   RE S UL T S   AND   DI S CU S S I ON     T he   s ys tema ti c   r e view   pr e s e nts   a   thor ough   s tudy   of   many   r e s e a r c h   s our c e s   to   a s s e s s   a nd   s ynthe s ize   inf or mation   a bout   f e de r a ted   DL   a nomaly   int r us ion   de tec ti on   in   S DN - ba s e d   I oT .   T he   f ind ings   c oll e c ted   indi c a te   a   va r iety   of   tec hniques   a nd   pr a c ti c e s   in   the   e xe c uti on   of   the   pr opos e d   methodology.   T his   s e c ti on   e xplains   the   r e s e a r c h   f indi ngs ,   whic h   of f e r   a   s umm a r y   of   the   pr e s e nt   s tudy.       5. 1.     S t at is t ical   m e t r ics   S e ve r a l   pe r f o r manc e   indi c a tor s   ha ve   be e n   de f ined   f or   the   mul ti - c las s   c onf us ion   matr ix   to   e va luate   the   e f f e c ti ve ne s s   of   the   model .   T he   mul ti - c las s   c onf us ion   matr ix   is   an   N   matr ix,   whe r e   N   r e pr e s e nts   the   number   of   unique   c las s   labe ls   ( C 0,   C 1,   ...,   C N) .   M a tr ix   c e ll s   a r e   de ter mi ne d   by   the   output   c ons is ti ng   of   the   pr e dicte d   labe l,   whic h   may   be   e it he r   pos it ive   or   ne ga ti ve ,   that   c omes   out   of   c ompar ing   the   pr e dic ted   labe l   with   the   a c tual   c las s   labe l,   whic h   can   be   e it he r   no r mal   or   a tt a c k   [ 42] .   As   a   r e s ult ,   the   tr a dit ional   c las s if ica ti on   of   tr ue   pos it ive   ( T P ) ,   tr ue   ne ga ti ve   ( T N) ,   f a ls e   pos it ive   ( F P ) ,   a nd   f a ls e   ne ga ti ve   ( F N)   c a s e s   be c omes   ir r e leva nt.   Al ter na ti ve ly,   a   mo r e   a ppr opr iate   a pp r oa c h   e ntails   f oc us ing   on   c e r tain   c las s e s .   T his   t e c hnique   a ll ows   f or   the   f o r mul a ti on   of   c las s - s pe c if ic   metr ics .   By   a de ptl y   mer ging   thes e   mea s ur e ments   that   a r e   dis ti nc t   to   each   c las s ,   a   c ompr e he ns ive   c oll e c ti on   of   metr ics   f or   the   whole   c onf us ion   matr ix   can   be   obt a ined,   as   e xe mpl if ied   in   ( 1)   to  ( 5 )   [ 43] .       =    ( )  = 1 ,  = 1  = 1   ( 1)     1 ( ) =   2 ( )  ( ) ( ) +  ( )   ( 2)      ( ) =    ( )  ( ) +  ( )   ( 3)   Evaluation Warning : The document was created with Spire.PDF for Python.
I nt  J   Ar ti f   I ntell     I S S N:   2252 - 8938       F e de r ated  de e lear ning  int r us ion  de tec ti on  s y s te on  s oft w ar e   de fi ne d - ne tw or k   …  ( He ba  Dhir ar )   3117    ( ) = ( 1 +   2 )   ( )  ( ) 2 ( ) +  ( )   ( 4)      ( ) =    ( )  ( ) +  ( )   ( 5)     5. 2.     E xp e r im e n t al   r e s u lt s   T he   im p leme ntation   of   the   DL   model   uti li z e d   T e ns or F low ,   Ke r a s ,   a nd   S c iki t - L e a r n   as   the   unde r lyi ng   tec hnology   a nd   wa s   e xe c uted   unde r   th e   gr a phics   pr oc e s s ing   unit   ( GPU )   T4   x2   e nvir onm e nt.   T he   pr oc e s s   of   FL   wa s   e xa mi ne d   f or   20   r ounds   the   e va luation   pa r a mete r s   f o r   each   r ound   a r e   il lus t r a ted   in     T a ble   6.   In   the   f ir s t   r ound,   only   one   c li e nt   wa s   us e d   whic h   obs e r ve d   a   h igh   de tec ti on   los s ,   r e a c hing   2. 8321   a nd   an   a c c ur a c y   of   0 . 0841.   T h is   can   be   a tt r ibut e d   to   the   li mi ted   dive r s it y   a nd   ins uf f icie nt   tr a in i ng   da ta.   C ompar a ti ve ly,   r unning   the   model   us ing   thr e e   c li e nts   f or   jus t   one   r ound,   de c r e a s e d   the   los s   to   0 . 0 987,   a nd   incr e a s e d   the   a c c ur a c y   to   0. 9749 .   F or   both   s c e na r ios ,   the   tes ts   we r e   c onduc ted   f or   50   e poc hs ,   wi th   a   ba tch   s ize   of   250   a nd   Ada m   opti m ize r   due   to   i ts   a da pti v e   lea r ning   r a te   f e a tur e s   a nd   dur a bil it y .   Af ter   c ompl e ti ng   the   20   tr a ini ng   r ounds ,   ther e   is   a   s igni f ica nt   e nha nc e ment   in   a c c ur a c y,   r is ing   f r om   99. 76   to   99 . 89% .   In   a d dit ion,   a   notable   r e duc ti on   de mons tr a ted   in   los s   de c r e a s e d   f r om   0. 01   in   the   s tanda r d   c e ntr a li z e d   tr a ini ng   pr oc e dur e   to   0. 005   in   the   f e de r a ted  DL   s c e na r io.   F igur e   5   il lus t r a tes   the   r e s ult s   of   each   c yc le,   with   F igur e   5 ( a )   s h owing   an   im pr ove ment   in   e nha nc e ment   a nd   F igur e   5 ( b)   indi c a ti ng   a   los s   r e duc ti on.       T a ble   6.   F e de r a ted   DL   tr a ini ng   r e s ult s   R ound   C li e nt   1   C li e nt   2   C li e nt   3     F e dA vg   A c c ur a c y   L os s   A c c ur a c y   L os s   A c c ur a c y   L os s   A c c ur a c y   L os s   1   0.9745   0.0986   0.9751   0.0984   0.9750   0.0984   0.9749   0.0987   2   0.9964   0.0131   0.9955   0.0142   0.9959   0.0136   0.9959   0.0139   3   0.9980   0.0093   0.9979   0.0093   0.9980   0.0093   0.9978   0.0094   4   0.9982   0.0090   0.9981   0.0089   0.9981   0.0090   0.9980   0.0091   5   0.9981   0.0089   0.9981   0.0086   0.9981   0.0088   0.9980   0.0088   6   0.9984   0.0078   0.9984   0.0077   0.9984   0.0078   0.9983   0.0079   7   0.9983   0.0079   0.9984   0.0077   0.9983   0.0079   0.9982   0.0079   8   0.9987   0.0068   0.9987   0.0068   0.9987   0.0069   0.9986   0.0069   9   0.9984   0.0073   0.9984   0.0071   0.9984   0.0072   0.9983   0.0073   10   0.9987   0.0069   0.9987   0.0069   0.9987   0.0070   0.9986   0.0070   11   0.9988   0.0064   0.9988   0.0064   0.9988   0.0064   0.9987   0.0065   12   0.9987   0.0066   0.9987   0.0066   0.9987   0.0066   0.9986   0.0067   13   0.9988   0.0060   0.9988   0.0060   0.9988   0.0060   0.9987   0.0061   14   0.9987   0.0067   0.9987   0.0068   0.9987   0.0068   0.9986   0.0069   15   0.9988   0.0060   0.9989   0.0061   0.9988   0.0061   0.9988   0.0061   16   0.9989   0.0057   0.9989   0.0057   0.9989   0.0058   0.9988   0.0058   17   0.9989   0.0056   0.9989   0.0056   0.9989   0.0057   0.9988   0.0057   18   0.9989   0.0059   0.9989   0.0059   0.9989   0.0059   0.9988   0.0060   19   0.9989   0.0059   0.9989   0.0059   0.9989   0.0059   0.9988   0.0060   20   0.9989   0.0057   0.9989   0.0057   0.9989   0.0057   0.9988   0.0057           ( a )   ( b)     F igur e   5.   C li e nt   t r a ini ng   met r ics   r e s ult s   f or   ( a )   mo de l   a c c ur a c y   metr ic   f o r   the   c li e nts   a nd   F e dAvg   a nd     ( b)   model   los s   metr ic   f or   the   c li e nts   a nd   F e dAvg     Evaluation Warning : The document was created with Spire.PDF for Python.
                                I S S N :   2252 - 8938   I nt  J   Ar ti f   I ntell Vol.   14,   No.   4,   Augus 2025 310 9 - 3120   3118   6.   CONC L USI ON   T his   wor k   pr e s e nts   a   f e de r a ted   DL   f or   NI DS   in   an   I oT   c ontext   in   a   mul ti - c ontr oll e r   S DN   ne twor k,   us ing   I P F S   as   the   unde r ly ing   tec hnology   a nd   the   AE S   e nc r ypt ion   a lgo r it hm   to   he lp   im pr ove   the   s e c ur it y   of   the   a ggr e ga ti on   a nd   tr a ini ng .   A   c us tom - ge ne r a ted   da tas e t   of   in tr a -   a nd   int e r - a tt a c ks   wa s   uti li z e d   to   e xtr a c t   int e r na l   f e a tur e   r e p r e s e ntations   to   de tec t   a nd   c las s if y   a tt a c ks .   T he   pr opos e d   a r c hit e c tur e   s uc c e s s f ull y   mi ti ga tes   DoS   a nd   DD oS   a tt a c ks   onc e   the   a tt a c k   t hr e s hold   is   r e a c he d   on   the   c ontr oll e r   to   a void   f loo ding   the   I DS   s e r ve r ,   whe r e   the   s ugge s ted   model   pos s e s s e s   an   a c c ur a c y   of   99. 89%   in   identif ying   s e ve r a l   a tt a c k   types   de mons tr a ted   s upe r ior   pe r f or manc e   in   both   the   d e tec ti on   a nd   c las s if ica ti on   of   a tt a c ks   u s ing   F L ,   s ur pa s s ing   c onve nti ona l   DL   f or   the   s a me   model .   T he   r e s ult   s hows   a   dr op   in   los s   f r om   0. 01   in   the   s tanda r d   c e ntr a li z e d   tr a ini ng   pr oc e dur e   that   u ti li z e s   the   DL   model   to   0. 005   in   the   f e de r a ted   DL   s c e na r io.   In   a ddi ti on,   t he r e   is   a   s igni f ica nt   e nha nc e ment   in   a c c ur a c y,   r is ing   f r om   99. 76   to   99. 89 % .   T he   s ugge s ted   tec hnique   can   a p ply   to   a   wide   r a nge   of   s it ua ti ons   a nd   may   be   include d   as   a   c omponent   in   a   r e a l - time   S DN - I oT   e nvi r on ment.   I ts   pur pos e   is   to   de tec t   a ny   a tt a c ks   a nd   c las s if y   them   int o   c e r tain   types ,   c a us ing   an   a lar m .   T he   c ur r e nt   wor k   is   s ubopti mal.   I ns tea d   of   us ing   a   method   that   s e lec ts   a ll   the   c ha r a c ter is ti c s ,   it   would   be   mor e   e f f e c ti ve   to   a pply   ke r ne l - ba s e d   m e thods   to   c hoos e   the   idea l   f e a tur e s .   T his   may   s igni f ica ntl y   e nha nc e   the   e f f e c ti ve ne s s   of   the   S DN - I oT   I DS .   In   a ddit ion,   doing   a   tho r ough   e xa mi na ti on   a nd   e va luation   of   the   model   a nd   a nothe r   model   ins ide   the   e nvir onment   is   c r uc ial,   as   the   major it y   of   ML   a nd   DL   models   a r e   s us c e pti ble   to   a dve r s a r ial   a tt a c ks .       F UN DI NG  I NF ORM AT I ON   Author s   s tate   no  f unding   invol ve d.       AU T HO CONT RI B U T I ONS   S T AT E M E N T     T his   jour na l   us e s   the  C ontr ibut o r   R oles   T a xo nomy  ( C R e diT )   to   r e c ognize   indi vidual   a uthor   c ontr ibut ions ,   r e duc e   a utho r s hip  dis putes ,   a nd  f a c il it a te  c oll a bor a ti on.       Nam e   of   Au t h or   C   M   So   Va   Fo   I   R   D   O   E   Vi   Su   P   Fu   He ba   Dhir a r                               Ali  H .   Ha mad                                 C     C onc e pt ua li z a ti on   M     M e th odol ogy   So     So f twa r e   Va     Va li da ti on   Fo     Fo r ma a na ly s is   I     I nve s ti ga ti on   R     R e s our c e s   D   :   D a ta  C ur a ti on   O   :   W r it in -   O r ig in a D r a f t   E   :   W r it in -   R e vi e w  &   E di ti ng   Vi     Vi s ua li z a ti on   Su     Su pe r vi s io n   P     P r oj e c a dmi ni s tr a ti on   Fu     Fu ndi ng a c qui s it io n         CONF L I CT   OF   I NT E RE S T   S T AT E M E N T   T he   a uthor s   de c lar e   that   they  ha ve   no   c onf li c ts   of   i nter e s r e late to  th is   wor k.       DA T AV AI L A B I L I T   T he   da ta  that   s uppor t   the  f indi ngs   of   thi s   s tudy  a r e   ope nly  a va il a ble  on   Ka ggle  a t   htt ps :/ /www . ka ggle. c om/ da tas e ts /heba dhir a r /s dn - i ot,   unde r   the  ti tl e   S DN - I oT   I ntr us ion   De tec ti on  Da tas e t.       RE F E RE NC E S   [ 1]   A.   A lr a w a is ,   A.   A lh ot ha il y,   C.   H u,   a nd   X.   C he ng,   F og   c omput in g   f or   th e   in te r ne t   of   th in gs :   s e c ur it y   a nd   pr iv a c y   is s ue s ,”   I E E E   I nt e r ne t   C om put in g ,   vol .   21,   no.   2,   pp.   34 42,   M a r .   2017,   doi :   10.1109/M I C .2017.37.   [ 2]   L.   L iu ,   B.   X u,   X.   Z ha ng,   a nd   X.   W u,   A n   in tr us io n   de t e c ti on   me th od   f or   in te r ne t   of   th in gs   ba s e d   on   s uppr e s s e d   f uz z y   c lu s te r in g,”   E ur as ip   J our nal   on   W ir e le s s   C o m m uni c at io ns   and   N e tw or k in g ,   vol .   2018,   no.   1,   2018,   doi :   10.1186/s 13638 - 018 - 1128 - z.   [ 3]   H.   Q us ht om   a nd   K.   R a ba ya h,   E nha nc in g   th e   Q oS   of   I oT   ne twor ks   w it h   li ght w e ig ht   s e c ur it y   pr ot oc ol   us in g   C ont ik i   O S ,”   I nt e r nat io nal   J our nal   of   C om put e r   N e tw o r k   and   I nf or m at io n   Se c u r it y ,   vol .   9,   no.   11,   pp.   27 35,   20 17,     doi :   10.5815/i jc ni s .2017.11.03.   [ 4]   Y.   L i,   R.   M a ,   a nd   R.   J ia o,   “A   hybr id   ma li c io us   c ode   de te c ti on   me th od   ba s e d   on   de e p   le a r ni ng,”   I nt e r nat io nal   J our nal   of   Se c ur it y   and   its   A ppl ic at io ns ,   vol .   9,   no.   5,   pp.   205 216,   2015,   doi :   10.14257/i js ia .2015.9.5.21.   [ 5]   J.   K one č ný,   H.   B.   M c M a ha n,   F.   X.   Y u,   P.   R ic ht á r ik ,   A.   T.   S u r e s h,   a nd   D.   B a c on,   F e d e r a te d   le a r ni ng:   s tr a te gi e s   f or   im pr ovi ng   c omm uni c a ti on   e f f ic ie nc y,”   ar X iv - C om put e r  S c ie nc e , pp. 1 - 10,   2016 .   [ 6]   P.   B ooba la n   et   al . ,   F us io n   of   f e de r a te d   le a r ni ng   a nd   in dus tr ia l   in te r ne t   of   th in gs :   a   s ur ve y,”   C om put e r   N e tw or k s ,   vol .   212,   2022,   doi :   10.1016/j .c omne t. 2022.109048.   Evaluation Warning : The document was created with Spire.PDF for Python.