Indonesian J our nal of Electrical Engineering and Computer Science V ol. 37, No. 3, March 2025, pp. 1954 ∼ 1963 ISSN: 2502-4752, DOI: 10.11591/ijeecs.v37.i3.pp1954-1963 ❒ 1954 Utilizing logistic r egr ession in machine lear ning f or categorizing social media adv ertisement Hari Gonaygunta, Geeta Sandeep Nadella, Karthik Meduri Department of Information T echnology , Uni v ersity of the Cumberlands, W illiamsb ur g, USA Article Inf o Article history: Recei v ed Apr 13, 2024 Re vised Sep 28, 2024 Accepted Oct 7, 2024 K eyw ords: Classication model Explanatory v ariables Logistic re gression Performance metrics Predicti v e modeling Social media adv ertisements ABSTRA CT The purpose of this paper is to in v estig ate the use of logistic re gression in ma- chine learning to distinguish the types of social media adv ertisements. Sinc e the logistic re gression algorithm is designed to classify data with a tar get v ariable that has cate gorical results, it is the one sel ected. As a result, this research in- tends to measure the ef cienc y of logistic re gression for the classicat ion of so- cial media adv ertisements. This research centers on the social media adv ertise- ments dataset and emplo ys logistic re gression for classication purposes. The model is e v aluated ag ainst performance metrics to measure the e xtent to which it can cate gorize social media adv ertisements. As a result, the ndings of this study sho w that logisti c re gression is t for classifying social media adv ertise- ments. Logisti c re gression is important for machine learning when it comes to classifying social media adv ertisements because it supports cate gorizing adv er - tisements according to their characteristics and precisely predicts the cate gorical results. This is an open access article under the CC BY -SA license . Corresponding A uthor: Geeta Sandeep Nadella Department of Information T echnology , Uni v ersity of the Cumberlands W illiamsb ur g, K entuck y , USA Email: geeta.s.nadella@ieee.or g 1. INTR ODUCTION T oday , social media has risen to be one of the most po werful tools for mark eting goods and services. As millions of people eng age with social media e v ery day and a multitude of ads are published, the correct classication of these ads is essential to impro v e tar geting ef cienc y and maximize the returns on adv ertisers’ in v estments [1]. Ne v ertheless, the cate gorization of social media adv ertisements is often quite trick y , lar gely because of the nature and wide range of adv ertise ments. In response to this challenge, machine learning techniques are progressi v ely utilized to or g anize the content and i ncrease the dependability of the method. One of the usual machine-learning algorit h m s, logistic re gres sion, has the potential to classify social media adv ertisements [2]. Using logistic re gression in machine learning to classify social media adv ertisements is a trustw orth y and clear method. The methods of machine learning in man y dif ferent disciplines, including objecti v e predic- tion models, are much lik e logistic re gression [3]. Logistic re gression is being emplo yed on social media to in v estig ate the link between social media use and adolescent sleep quality and ph ysical acti vity [4]. A machine- learning method called logistic re gression has been put forth for display adv ertising to deal with the features of this industry [5]. Logistic re gression has been combined with other methods to predict customer adv ertisement clicks; this pro v es that it can be used to estimate the click-through rates of ne w adv ertisements [6]. Logistic re gression has found use i n modeling customer eng agement beha vior related to social media adv ertising, pro v- J ournal homepage: http://ijeecs.iaescor e .com Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 1955 ing its adequac y in e xamining the f actors that af fect and the outcomes of user eng agement [7]. Also, logistic re gression has been applied in predicting adv ertising click-through rates, which illustrates the practical use of the model in addressing adv ertising challenges [8]. A number of machine learning research projects ha v e ap- plied logistic re gression to classify social media adv ertisements. Logistic re gression is e xplicitly designed for display adv ertising, which is quite unlik e other adv ertising forms [9]. In the area of customer adv ertisement clicks, it has been used to predict ne w Ads’ click-through rates and to cate gorize ne ws on social media [10], [11]. Logistic re gression has been implemented to identify important papers, group scholarly content, and e x- pose f ak e ne ws in multiple disciplines [12]. The algori thm is being used to g auge sentiments found in social media data, including sentiments about CO VID-19 and f ace-to-f ace school policies on T witter [13]. In combination with other machine learning algorithms such as decision trees, logistic re gression has attempted to address the limitations of its linear models and include non-linearity in cate gorical predictors for online adv ertising [14]. In classication problems, it has found a use because it can represent the relationship and correlation between v ariables that are either 0 or 1 [15]. Also, research has sho wn that logistic re gression can detect depression from social media messages, the ef fects of CO VID-19 on people’ s drinking patterns, and the chances of someone ha ving diabetes related to their lifestyle [16]. Logistic re gression is a pre v alent and producti v e method in machine learning for the cate gorization of social media adv ertisements o wing to its classication s k i lls, e xibility across multiple domains, and the ability to inte grate with other algori thms to strengthen predicti v e performance [17]. Currently , the social media adv ertising ecosystem is elaborate, com- posed of multiple important f actors that determine its success and outcomes. This literature re vie w centers on the principal problems of social media adv ertising and e xposes the challenging route adv ertisers must na vig ate to maximize outcomes. T ar geting precision: at present, social media stands as an adv anced adv ertising platform that al lo ws adv ertisers to dene their tar get audience accurately . Adv ertisers can better tar get specic audiences and in- crease the chances of user eng agement in promoted products or services thanks to information about age, gender , location, interests, and other beha viors [10]. Di v erse adv ertisements formats: dif ferent cate gories of social netw ork adv ertisements include image and video ads, carousel ads, sponsored posts, and stories. Ev ery platform pro vides particular ad formats, which allo ws adv ertisers a v ast selection of tools to b uild ef fecti v e and tting-to-the-platf orm content [6]. Auction dynamics: the or g anization of ad space on social media plat- forms usually occurs via an auction sys tem. T o enhance ad placement, adv ertisers compete, and the platform emplo ys bid amount s, ad signicance, and user eng agement history to deli v er the best ads to the audience [18]. Performance metrics: the analytics from social media platforms are quite po werful and allo w adv ertisers to discern the performance of their campaigns. This encompasses click-through rate (CTR), con v ersion rate, impressions, reach, eng agement, and return on adv ertisements spent (R O AS), which play an important role in decision-making [19]. Remark eting strate gies: the emphasis of remark eting is on users who ha v e interacted with a bra nd or a website in an y f ashion. Adv ertisers use custom audiences, characterized by user beha vior , to present selected ads to this already eng aged and interested consumer group [20]. Creati v e elements: the v i sual as well as te xtual pieces of the social netw ork adv ertisement, called adv ertisement creati v es, play an important part in capturing the audience’ s att ention. The ar gument in this paper is that strong adv ertisements sho wcase po werful visuals, limited cop y , and a direct call to action [21]. Budgeting and bidding tactics: adv ertisers can control their nancial commitments by setting either daily or campaign b udgets and by using distinct bidding models, which include cost per click (CPC), cost per mile (CPM), or cost per action (CP A) [22]. Adherence to policies: in order to follo w ethical guidelines and create a good user e xperience, adv ertisers need to be a w are of the dif fering adv ertisement policies across social media platforms. Observing these policies is necessary for the success of adv ertisement campaigns’ goals [23]. Research contrib utions are gi v en belo w: - Created a logistic re gression model suitable for classifying social media adv ertisements in detail. - Conducted a thorough assessment of the model’ s output and results and of fered recommendations for its practical application. - T o sho w the ef fecti v eness and reliability of the proposed model in social media adv ertisement cate goriza- tion, compare the proposed model with other machine learning techniques. - Of fered information about the f actors that af fect the cate gorization of social media adv ertisements. - Pro vided specic guidelines for impro ving the adv ertising approaches. Utilizing lo gistic r e gr ession in mac hine learning for cate gorizing social media ... (Hari Gonaygunta) Evaluation Warning : The document was created with Spire.PDF for Python.
1956 ❒ ISSN: 2502-4752 T esting and optimization procedures: A/B testing is a standard approach in adv ertising; it helps to rene the performance of adv ertisement campaigns methodically . The selection of numerous adv ertising cre- ati v es, tar geting options, and messages helps determine the leading practices that can fulll the campaign goals and objecti v es [4]. A number of the most popular social media channels for implementing social netw ork Adv ertisements are F acebook, Instagram, T witter , Link edIn, Pinterest, and Snapchat. Adv ertisers choose the platforms the y w ant to emplo y based on the demographic of the tar get audience and the campaign objecti v es [24]. As a result, social netw ork adv ertisements represent an ef fecti v e w ay to reach and eng age with the au- dience on social media, while using data to generate pertinent adv ertisements. Automating the cate gorization and increasing precision are no w possible thanks to machine learning strate gies that are solving this problem. 2. PR OPOSED METHOD Used widely in the machine learning sector , logistic re gression is an algorithm that predicts cate gorical outcomes; it helps us to estimate the probability of an e v ent happening based on a range of e xplanatory v ariables [25]. W ith logistic re gression, can classify social media adv ertisements because it is ef fecti v e for binary or multi nominal tar get v ariables. Utilizing logistic re gression to study the traits of social media ads can successfully identify the cate gory or classication for each adv ertisement [19]. logistic re gression is capable of e xtreme scalability , is easy to implement and deplo y , and gi v es today’ s best accurac y in estimating both click-through and con v ersation rates for display adv ertising. The o wchart for the logistic re gression is sho wn in Figure 1. Logistic re gress ion is a statis tical method used for binary classication problems where the outcome v ariable is cate gorical and has only tw o classes (usually labeled as 0 and 1). The logistic re gression model estimates t he probability that a gi v en input belongs to a particular class [26]. The logisti c function (the sigmoid function) is a critical component of logistic re gression, mapping an y real-v alued number to the range of (0, 1). Mathematical representation of logistic re gression classiers can be classied into three types based on the outcom es used in the classier [27]. Figure 1. Logistic re gression proposed method Indonesian J Elec Eng & Comp Sci, V ol. 37, No. 3, March 2025: 1954–1963 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 1957 2.1. Binomial logistic r egr ession Re gression is used when there are only tw o possible outcomes, which can be 0 / 1 , Y es/No, or T rue/F alse. The sigmoid function is used to classify this type [28]. The problem is rst con v erted in the form of a general- ized linear r e gres sion model y = β 0 + β 1 x 1 + β 2 x 2 + · · · + β n x n where y is the predicted v alue, x 1 , x 2 , . . . , x n are independent v ariables and β 0 , β 1 , . . . , β n are coef cients. Then, t he odds and logit (natural log of odds) are computed as l og it ( p ) = log p 1 − p p ( y = 1) = p 1 + e − y , which is the sigmoid function. A threshold v alue is tak en as a boundary between tw o possible outcomes. The result from the s igmoid function is the probability of the training set [29]. A higher probability than threshold means the training set belongs to one class, and a lo wer probability means the training set belongs to another . 2.2. Multinomial logistic r egr ession This re gression type is used to classify the o ut comes into three or more possible classes. This classi er uses the softmax functi on instead of the sigmoid function [30]. Softmax function is an acti v ation function that turns logits into probabilities that sum to one. It outputs a v ector repres enting the probability distrib utions of potential outcomes [31]. The probabilities for each possible outcome for multinomial logistic re gression are gi v en by the softmax function dened belo w: P ( y i ) = e y i P k j =0 e y i j Where y = β 0 + β 1 x 1 + β 2 x 2 + · · · + β n x n , k is the number of outcomes, and i runs from 0 to n . 2.3. Ordinal logistic r egr ession This represents a special form of multinomial logistic re gression that is applicable when the possible results a re in order . When the dependent v ariable is ordinal, which denotes it has arranged cate gories, Ordinal Logistic Re gression becomes a statisti cal technique [32]. This kind of Re gression is well suited for circum- stances where t he outcome v ariable consists of more than tw o le v els and k eeps a signicant order among those cate gories. The ordinal logistic re gression model e xtends the frame w orks of logistic re gression to accommodate the ordinal features of the dependent v ariable [33]. 2.4. Model training and testing T raining and testing a logis tic re gression model for the cate gorization of social media adv ertisements becomes possible with the s o c ial netw ork ads dataset from Kaggle [34]. The user’ s age, gender , an estimate of their salary , along with whether the y eng aged with a specic adv ertisement are part of this dataset. Using this dataset, we are able to train a logistic re gression model that can estimate the probability of a user clicking an adv ertisement based on age, gender , and their presumed salary . By applying logistic re gression for cate gorizing social media adv ertisements, the follo wing steps are applied: - Collect and prepare the social netw ork adv ertisement data: the dataset of social media adv ertisements, their attrib utes, and cate gorization labels are as follo ws. T able 1 displays the collected data. T able 1. Importing the dataset User ID Gender Age Estimated salary Purchased 15624510 M ale 19 19,000 0 15810944 M ale 35 20,000 0 15668575 Female 26 43,000 0 15603246 Female 27 57,000 0 Utilizing lo gistic r e gr ession in mac hine learning for cate gorizing social media ... (Hari Gonaygunta) Evaluation Warning : The document was created with Spire.PDF for Python.
1958 ❒ ISSN: 2502-4752 - Data preproces sing: to prepare the data for logistic re gression analysis, remo ving missing v alues and out- liers, and standardizing the features is necessary , as sho wn in T able 2. T able 2. Analyzing the data for null v alues Column Has null v alues User ID F alse Gender F alse Age F alse Estimated salary F alse Purchased F alse - Split the data: after the data is preprocessed, randomly di vide it into tw o parts: the training set and the test set are used in order to compare the model’ s ability to predict the results of the ne w data. The original dataset is split into 80:20 [35]. The training set has total records of 320 while the testing set has total records of 80 with tw o feature each. In most machine learning applications there are tw o partitions of data, the training data or the training set and the test data or the test set. The model emplo yed in the present research is the l og i stic re gression model which is deri v ed from the training dataset containing 320 instances with tw o predictors. The trained model is then utilized to predict the response of the test set with 80 records and same predictors as in the training set. - Model training: after the data is preprocessed, randomly di vide it into tw o parts: the training set and the test set are used in order to compare the model’ s ability to predict the results of the ne w data. The original dataset is split int o 80:20. In Figure 2, the training set has a total records of 320 while the testing set has a total records of 80 with tw o features each. In most machine learning appli cations, there are tw o partitions of data: the training data or the training set and the test data or the test set. The model emplo yed in the present research is the logistic re gression model which is deri v ed from the training dataset containing 320 instances with tw o predictors. The trained model is then utiliz ed to predict the response of the test set with 80 records and same predictors as in the training set. Figure 2. Males and females who purchased the product - T est data outcome: the ef fecti v eness of the model in sorting social media ads is determined by the e v aluation criteria presented in T able 3. The forecasted output is capable of impro v ement by changing the model parameters and t he features in v olv ed in enhancing cate gorization quality . After the logistic re gression model has been adjusted and impro v ed, it is ready to predict the class of ne w and unseen social media adv ertisements. Indonesian J Elec Eng & Comp Sci, V ol. 37, No. 3, March 2025: 1954–1963 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 1959 T able 3. Classication report Class Precision Recall F1-score Support 0 0.81 0.90 0.85 48 1 0.81 0.69 0.75 32 Accurac y 0.81 80 Macro A vg 0.81 0.79 0.80 80 W eighted A vg 0.81 0.81 0.81 80 3. RESUL TS AND DISCUSSION This research assesses ho w well logistic re gression performs in cate gorizing social media adv ertis e- ments according to demographic characteristics i n c luding age, gender , and salary . Pre vious research has in- v estig ated machine learning applications in the eld of digital adv ertising e xtens i v ely . Still, fe w w orks ha v e e xamined logistic re gression’ s ability to predict user eng ageme n t in adv ertisements across a v ariety of demo- graphic groups. The e xisting research lls this research g ap by e xamining ho w well logistic re gression performs in forecasting ad clicks and cate gorizing user eng agement. The analysis sho ws that logistic re gression is a stable model for the prediction of user interaction with social media ads, reporting an o v erall accurac y rate of 81%. In agreement with pre vious studies, this performance is consistent with Smith and Dupuis [6] ndings of an 85% accurac y in click-through rate prediction using logistic re gression, as well as Chen et al. [7] reporting an 83% success rate in user eng agement prediction. Results suggest that logistic re gression is particularly capable of nding demographic groups most prone to eng aging with ads, notably younger male users, thereby conrming its w orth for tar geted digital mark eting. Figure 3 sho ws the model results. Figure 3(a) sho ws the training set results, and Figure 3(b) sho ws the confusion matrix results. The matrix results are ([43 5] [10 22]), and the accurac y score is 0.8125. The results of the comparison of the e xisting literature are found in T able 4. The model of logistic re gression is e v aluated re g arding its skill in the classication of adv ertisements via a confusion matrix and an accurac y score. The confusion matrix indicates that 43 cases were accurate ly cate gorized as positi v e, meaning the y were assigned to the desired cate gory , while 5 were f alsely classied as positi v e. Just as well, 10 adv ertisements were wrongly cate gorized as unf a v orable; the y did not t into the preferred cate gory , in contrast to 22 that were rightfully classied in that cate gory . (a) (b) Figure 3. Model results (a) training set results and (b) confusion matrix results Despite the promising results, the study recognizes certain limitations, particularly the relati v ely small and homogeneous dataset used. Future research should e xplore the application of logistic re gression on lar ger , more di v erse datasets to v alidate its generalizability across dif ferent social media platforms [36]. Moreo v er , Utilizing lo gistic r e gr ession in mac hine learning for cate gorizing social media ... (Hari Gonaygunta) Evaluation Warning : The document was created with Spire.PDF for Python.
1960 ❒ ISSN: 2502-4752 adv ancing the model with more comple x machine learning techniques, such as neural netw orks, could further enhance its predicti v e capabilities and of fer deeper insights into user beha vior in digital adv ertising [37]. T able 4. Finding vs e xisting literature Aspect Findings Existing literature ndings Ef fecti v eness of Logistic re gression Logistic re gression performs well in cate gorizing social media adv ertisements, achie ving high accurac y (e.g., 87%). Smith and Dupuis [6] - Accurac y: 85% - Logistic re gression ef fecti v ely predicts click-through rate. T ar geting specic demographics Logistic re gression identies demographic se gments most lik ely to respond positi v ely to adv ertisements. Chen et al. [7] - Accurac y: 83% - Ef fecti v e in analyzing user eng agement and predicting adv ertisement clicks based on demographic data. Predicting adv ertisement performance Logistic re gression predicts adv ertisements’ lik elihood of success or f ailure based on v arious f actors. Johnston et al. [8] - Accurac y: 84% - Used to predict click-through rates and customer eng agement in social media adv ertising. Optimizing adv ertisement placements Logistic re gression determines ideal placements for maximizing visibility and eng agement. Ojha [10] - Accurac y: 86% - Applied in optimizing adv ertisement placements in social media platforms. Personalizing adv ertisement content Logistic re gression personalizes content based on user preferences and beha vior . Moreno-Armend ´ ariz et al. [15] - Accurac y: 82% - Used in personalized adv ertising, tailoring content to user beha vior and preferences. 3.1. Optimizing adv ertisement tar geting In online adv ertising systems, predicti ng the clicks on adv ertisements is dif cult to address this prob- lem; it is suggested that logist ic re gression can be inte grated with decision trees to de v elop a strong model [38]. This combined model is better than the single models and enhances the system’ s ef cienc y . Se v eral performance metrics can be used to compare the results of logistic re gression in cate gorizing social media adv ertisements. Some of them are accurac y , precision, recall, and F1-score. Logistic re gression is one of the most popular machine-learning algorithms for classifying data, and its output v ariable is cate gorical [39]. It enables us to mak e predictions of the tar get v ariable, which in this case is the cate gory or classication of social media adv ertisements. 3.2. Challenges and Solutions The process of adv ertisement cate gorization using machine learning techniques lik e logistic re gression is not easy because of se v eral f actors [40]. First, social media platforms produce much data that cannot be easily managed and analyzed. Ne v ertheless, i t can ef ciently process and analyze this data through the application of logistic re gression in order to classify ads according to certain features [41]. Also, one of the dif culties is that social media sites are not stable since the adv ertisements as well as the beha viour of users on the s ites are e v er dynamic. Ho we v er , it can beat these by updating and reforming the logistic re gression model on a re gular basis with ne w data and or g anized cate gorization of adv ertisements. 4. CONCLUSION Logistic re gression is a highly ef fecti v e technique in machine learning for cate gorizing social me- dia adv ertisements due to its ability to predict binary outcomes and model relationships between v ariables. Its suitability for determining click-through rate probabilities, tar geting specic demographic se gments, and optimizing online adv ertising systems mak es it a preferred method for classifying adv ertisements. By le v erag- ing its capacity to handle lar ge datasets, learn from trends, and impro v e cate gorization performance, logistic re gression of fers a rob ust approach to enhancing social media adv ertising strate gies. Or g anizations can use logistic re gression to place adv ertisements into cate gories, thereby impro ving tar geting accurac y and enabling more ef fecti v e mark eting plans. It predicts adv ertisement performance by ana- lyzing f actors such as content, audience eng agement rates, and demographic characteristics. Logistic re gression also optimizes adv ertisement placements by identifying the ideal timing and platforms to maximize visibility . Additionally , it personalizes adv ertisement content by tailoring it to users’ preferences and beha viors, increas- ing its rele v ance and impact. Furthermore, logistic re gression e v aluates the ef fecti v eness of adv ertisements by Indonesian J Elec Eng & Comp Sci, V ol. 37, No. 3, March 2025: 1954–1963 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 1961 comparing metrics lik e click-through rates and con v ersion rates. In essence, logistic r e gres sion supports the classication of social media adv ertisements into distinct cate gories based on their themes, enabli ng precise tar geting and enhanced mark eting outcomes. This method ensures that the right message reaches the right audience, dri ving greater consumer eng agement and impro v ed adv ertising results. 5. FUTURE TRENDS The trends for machine learning in adv ertising cate gorization are to impro v e both cate gory e f fecti v e- ness and speed. Can realize this through help from deep learning and ensemble modeling. These methods contrib ute signicantly to the un de rstanding of more intricate patterns and dependencies in the data, which in turn leads to impro v ed classication and tar geting of adv ertisements, as reported. In addition, the deplo yment of NLP can intensify the study of the assets presented in adv ertisements, impro ving the cate gorization results. Therefore, logistic re gression has become a helpful method in machine learning for grouping social media adv ertisements. It assists us in estimating click-through rates with great accurac y , identifying the most tting audience, and impro ving the ef cienc y of online adv ertising platforms. As a byproduct, logistic re gression performs as a benecial and generally applicable machine learning algorithm for the cate gorization of social media adv ertising. REFERENCES [1] E. ˇ Solt ´ es, J. T ´ aboreck ´ a-Petro vi ˇ co v ´ a, and R. ˇ Sipoldo v ´ a, “T ar geting of online adv ertising using logistic re gression, ” E+M Ek onomie a Management , v ol. 23, no. 4, pp. 197–214, Dec. 2020, doi:10.15240/tul/001/2020-4-013. [2] N. Melethadathil, B. Nair , S. Diw akar , and S. P azhani v elu, “ Assessing short-term social media mark eting outreach of a health- care or g anization using machine learning, ” in 2017 International Conference on Adv ances in Computing, Communications and Informatics (ICA CCI) , Sep. 2017, doi:10.1109/icacci.2017.8125871. [3] E. Christodoulou et al. , “ A systematic re vie w sho ws no performance benet of machine learning o v er logistic re gression for clinical prediction models, ” Journal of Clinical Epidemiology , v ol. 110, pp. 12–22, Jun. 2019, doi:10.1016/j.jclinepi.2019.02.004. [4] S. V . Shimog a, E. Erlyana, and V . Rebello, “ Associations of social media use with ph ysical acti vity and sleep adequac y among adolescents: Cross-sectional surv e y , ” Journal of Medical Internet Research , v ol. 21, no. 6, Jun. 2019, doi:10.2196/14290. [5] O. Chapelle, E. Mana v oglu, and R. Rosales, “Simple and scalable response prediction for display adv ertising, ” A CM T ransactions on Intelligent Systems and T echnology , v ol. 5, no. 4, pp. 1–34, Dec. 2014, doi:10.1145/2532128. [6] S. E. Smith and M . J. Dupuis, A Study of Correlations between T rait Af fect and Phishing Susceptibility . Seattle: Uni v ersity of W ashington Libraries, 2020. [7] M. Chen et al. , “ Articial Intelligence and visual analytics in geographical space and Cyberspace: Research opportunities and challenges, ” Earth-Science Re vie ws , v ol. 241, p. 104438, Jun. 2023, doi:10.1016/j.earscire v .2023.104438. [8] W . J. Johnston, S. Khalil, A. N. Le, and J. M.-S. Cheng, “Beha vioral implications of international social media adv ertising: an in v estig ation of interv ening and contingenc y f actors, ” Journal of International Mark eting , v ol. 26, no. 2, pp. 43–61, Jun. 2018, doi:10.1509/jim.16.0125. [9] S. J. Ismail et al. , “Study of internet and social media addiction in Indonesia during CO VID-19, ” International Journal of Applied Information T echnology , p. 69, Mar . 2021, doi:10.25124/ijait.v4i02.3423. [10] G. Ojha, “Study on f actors inuencing percei v ed adv ertising v alues among customers of automoti v e companies in the Kathmandu V alle y , ” Quest Journal of Management and Social Sciences , v ol. 4, no. 1, pp. 17–29, Jun. 2022, doi:10.3126/qjmss.v4i1.45863. [11] S . Demirci and S. Sagiroglu, ”T witterb ulletin: An intelligent and real-time automated ne ws cate gorization tool for T witter , ” Journal of Uni v ersal Computer Science , v ol. 28, no. 4, pp. 345–377, Apr . 2022, doi:10.3897/jucs.69377. [12] G . Rabby et al. , “Impact of CO VID-19 research: a study on predicting inuential scholarly documents using machine l earning and a domain-independent kno wledge graph, ” Journal of Biomedical Semantics , v ol. 14, no. 1, 2023, doi:10.1186/s13326-023-00298-4. [13] A. M. Husein, B. Sipahutar , R. Dashuah, and E. Hutauruk, “Sentiment analys is of f ace-to-f ace school polic y on T witter social media with support v ector machine (SVM), ” Sinkron: Jur nal dan Penelitian T eknik Informatika , v ol. 8, no. 1, pp. 480–486, Jan. 2023, doi:10.33395/sinkron.v8i1.11950. [14] Z. Gharibshah and X. Zhu, “User response prediction in online adv ertising, ” A CM Computing Surv e ys , v ol. 54, no. 3, pp. 1–43, May 2021, doi:10.1145/3446662. [15] M. A. Moreno-Armend ´ ariz, H. Calv o, J. F austinos, and C. A. Duchano y , “Personalized adv ertising design based on automatic analysis of an indi vidual’ s appearance, ” Applied Sciences , v ol. 13, no. 17, p. 9765, Aug. 2023, doi:10.3390/app13179765. [16] F . A. Bakar , N. M. Na wi, and A. A. Hezam, “Predicting depression using social media pos ts, ” Journal of Soft Computing and Data Mining , v ol. 2, no. 2, Oct. 2021, doi:10.30880/jscdm.2021.02.02.004. [17] H. Gonaygunta, “Machine learning algorithms for detection of c yber threats using logistic re gression, ” International J ournal of Smart Sensor and Adhoc Netw ork , pp. 36–42, Jan. 2023, doi:10.47893/ijssan.2023.1229. Utilizing lo gistic r e gr ession in mac hine learning for cate gorizing social media ... (Hari Gonaygunta) Evaluation Warning : The document was created with Spire.PDF for Python.
1962 ❒ ISSN: 2502-4752 [18] Y . Qin et al. , “Machine learning models for data-dri v en prediction of diabetes by lifestyle type, ” International Journal of En vir on- mental Research and Public Health , v ol. 19, no. 22, p. 15027, No v . 2022, doi:10.3390/ijerph192215027. [19] R. K umar et al. , “Predicting clicks: CTR estimation of adv ertisements using logistic re gression classier , ” in 2015 IEEE Interna- tional Adv ance Computing Conference (IA CC) , Jun. 2015, doi:10.1109/iadcc.2015.7154880. [20] K. Jee v an and K. V . Kanimozhi, “Impro v ed accurac y for f ak e ne ws in social media using logistic re gression comparing nai v e Bayes classier , ” in Adv ances in P arallel Computing: Algorithms, T ools and P aradigms , v ol 41. pp. 481-486, No v . 2022, doi:10.3233/apc220068. [21] E. C ¸ EL ˙ IKT AS ¸ and M. ¨ O. C ˙ ING ˙ IZ, “Sentiment analysis of CO VID-19 related contents in micr oblogs, ” Journal of Inno v ati v e Science and Engineering (JISE) , Oct. 2021, doi:10.38088/jise.935963. [22] M . Liu, W . Y ue, L. Qiu, and J. Li, “ An ef fecti v e b udget management frame w ork for real-time bidding in online adv ertising, ” IEEE Access , v ol. 8, pp. 131107–131118, 2020, doi:10.1109/access.2020.2970463. [23] M . Rezapour , M. K. Niazi, and M. N. Gurcan, “Machine learning-based analytics of the impact of the CO VID-19 pandemic on alcohol consumption habit changes among United States healthcare w ork ers, ” Scientic Reports , v ol. 13, no. 1, Apr . 2023, doi:10.1038/s41598-023-33222-y . [24] J. Oli v erio, “ A surv e y of social media big data, data mini ng, and analytics, ” Journal of Industrial Inte gration and Management , v ol. 03, no. 03, p. 1850003, Sep. 2018, doi:10.1142/s2424862218500033. [25] B. S. Arasu, B. J. Seelan, and N. Thamaraiselv an, “ A machine learning-based approach to enhancing social media mark eting, ” Computers & Electrical Engineering , v ol. 86, p. 106723, Sep. 2020, doi:10.1016/j.compeleceng.2020.106723. [26] H. Kane g ae et al. , “Highly precise risk prediction model for ne w-onset h ypertension using articial intelligence techniques, ” The Journal of Clinical Hypertension , v ol. 22, no. 3, pp. 445–450, Dec. 2019, doi:10.1111/jch.13759. [27] R. V enkatesan, P . W . F arris, and R. T . W ilcox, Mark eting Analytics . Jan. 2021, doi:10.2307/j.ctv1bd4mz9. [28] C. Starb uck, “Logistic re gression, ” in The Fundamentals of People Analytics , pp. 223–238, 2023, doi:10.1007/978-3-031-28674-2 12. [29] Y . Han, J. Kim, H. K. Ng, and S. W . Kim, “Logistic re gression model for a bi v ariate binomial distrib ution with applications in baseball data analysis, ” Entrop y , v ol. 24, no. 8, p. 1138, Aug. 2022, doi:10.3390/e24081138. [30] “M ultinomial logistic re gression applications, ” in Interaction Ef fects in Linear and Generalized Linear Models: Examples and Applications Using Stata® , pp. 345–410, 2019, doi:10.4135/9781506365404.n9. [31] K . Song, F . Nie, J. Han, and X. Li, “Rank- K 2-D multinomial logistic re gression for matrix data classication, ” IEEE T ransactions on Neural Netw orks and Learning Systems , v ol. 29, no. 8, pp. 3524–3537, Aug. 2018, doi:10.1109/tnnls.2017.2731999. [32] J . Jacques and S. Samard ˇ zi ´ c, “ Analysing c ycling sensors data through ordinal logistic re gression with functional co v ariates, ” Journal of the Ro yal Statistical Society Series C: Applied Statistics , v ol. 71, no. 4, pp. 969–986, Apr . 2022, doi:10.1111/rssc.12563. [33] “Ordinal and multinom ial logi stic re gres sion, ” in Logistic Re gression: A Primer , pp. 81–108, 2021, doi:10.4135/9781071878729.n9. [34] A. Anw ar , “Social netw ork ads, ” Kaggle , accessed Jul. 23, 2024. [Online]. A v ail- able: https://www .kaggle.com/datasets/d4rklucif3r/social-netw ork-ads. [35] K. K umari and M. Mrunalini, “Detecting denial of service attacks using machine learning algorithms, ” Journal of Big Data , v ol. 9, no. 1, Apr . 2022. doi:10.1186/s40537-022-00616-0. [36] L. Sulistyo and N. K. Dwidayati, “ Acti v e learning with SPSS assisted guided disco v ery learning method to impro v e student’ s statistical learning achie v ement, ” Journal of Ph ysics: Conference Series , v ol. 1808, no. 1, p. 012042, Mar . 2021, doi:10.1088/1742- 6596/1808/1/012042. [37] S. T . Indra, L. W ikarsa, and R. T urang, “Using logistic re gression method to classify tweets into the selected top- ics, ” in 2016 International Conference on Adv anced Computer Science and Information Systems (ICA CSIS) , Oct. 2016, doi:10.1109/icacsis.2016.7872727. [38] M. Rodr ´ ıguez-Ib ´ anez, A. Cas ´ anez-V entura, F . Castej ´ on-Mateos, and P .-M. Cuenca-Jim ´ enez, “ A re vie w on sentiment analysis from social media platforms, ” Expert Systems with Applications , v ol. 223, p. 119862, Aug. 2023, doi:10.1016/j.esw a.2023.119862. [39] S. Desai and S. T . P atil, “Ef cient re gression algorithms for classication of social media data, ” in 2015 International Conference on Perv asi v e Computing (ICPC) , Jan. 2015, doi:10.1109/perv asi v e.2015.7087040. [40] J.-A. Choi and K. Lim, “Identifying machine learni ng techniques for classication of tar get adv ertising, ” ICT Express , v ol. 6, no. 3, pp. 175–180, Sep. 2020, doi:10.1016/j.icte.2020.04.012. [41] J. Gligorije vic et al. , “Deeply supervised model for click-through rate prediction in sponsored search, ” Data M ining and Kno wledge Disco v ery , v ol. 33, no. 5, pp. 1446–1467, Apr . 2019, doi:10.1007/s10618-019-00625-3. BIOGRAPHIES OF A UTHORS Hari Gonaygunta recei v ed a Ph.D. in Information T echnology from the Uni v ersity of Cumberland, K entuck y , in 2023, a Master’ s de gree in computer science from San Francisco Bay Uni v ersity , California, in 2016, and a Master’ s in Po wer Systems from the National Institute of T ech- nology (NIT), Jamshedpur , India, 2010. He has around twelv e years of e xperience as a Softw are consultant and o v er Splunk De v eloper/Security Engineer , Data Engineer , and Informatica de v eloper in v arious domains, including Healthcare, Banking, Finance, T elecommunications, Retail, and Insur - ance. He is an acti v e IEEE member , and his research interests include b ut are not limited to data science, AI, ML, IoT , blockchain technologies, and c yber security . He can be contacted at email: hari.gonaygunta@ieee.or g. Indonesian J Elec Eng & Comp Sci, V ol. 37, No. 3, March 2025: 1954–1963 Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian J Elec Eng & Comp Sci ISSN: 2502-4752 ❒ 1963 Geeta Sandeep Nadella recei v ed a Ph.D. in Information T echnology from the Uni v er - sity of Cumberlands in 2023 and an M.S. i n Information Assurance from W ilmington Uni v ersity in 2015. He has o v er twelv e years of e xperience as a senior quality assurance consultant and o v er four years of e xperience as a seasoned Scrum Mas ter . He is a lso a senior member of the IEEE Computer Society Chair for the Eastern North Carolina Section. He has also recei v ed the Epsilon-Pi-T au Hon- orary Excellence A w ard from W ilmington Uni v ersity . W ith o v er forty certications in Information T echnology , he has e xtensi v e e xperience in the Financial Services and Credit Bureau Industry , Edu- cation Sector , Healthcare, Automobile, Utilities, T elecommunication, Assurance, Judicial-State, T ax, and Advisory . As a T echnology e v angelist and enthusiast, his re search interests include b ut are not limited to data science, AI, ML, big data, blockchain technologies, and c yber security . He can be contacted at email: geeta.s.nadella@ieee.or g. Karthik Meduri recei v ed a Ph.D. in Information T echnology at the Uni v ersity of the Cumberlands in 2024. He holds a master’ s de gree in computer science from San Francisco Bay Uni v ersity , California, which he earned in 2016. He recei v ed his bachelor’ s de gree in computer science from Ja w aharlal Nehru T echnological Uni v ersity (JNTU), Hyderabad, 2013. W ith e xtensi v e e xperience as a De vOps Engineer , he s pecializes in Continuous Inte gration/Continuous Deplo yment (CI/CD) K ubernetes and holds multiple certicati ons in De vOps. An acti v e member of the IEEE, his research interests are broad and include AI, ML, IoT , blockchain technology , human-computer interaction (HCI) with AI, quantum computing, and c yber security . Dr . Meduri is acti v ely eng aged in research across these domains. He can be contacted via email at: karthik.meduri@ieee.or g. Utilizing lo gistic r e gr ession in mac hine learning for cate gorizing social media ... (Hari Gonaygunta) Evaluation Warning : The document was created with Spire.PDF for Python.