TELK OMNIKA T elecommunication, Computing, Electr onics and Contr ol V ol. 24, No. 2, April 2026, pp. 527 ∼ 535 ISSN: 1693-6930, DOI: 10.12928/TELK OMNIKA.v24i2.27240 ❒ 527 Hybrid classical–quantum ensemble lear ning f or r eal-time ight delay pr ediction at T ribhuv an Inter national Air port P a v an Khanal 1 , Nanda Bikram Adhikari 2 1 Ci vil A viation Authority of Nepal, Kathmandu, Nepal 2 Department of Electronics and Computer Engineering, IOE Pulcho wk Campus, T ribhuv an Uni v ersity , Lalitpur , Nepal Article Inf o Article history: Recei v ed Jun 16, 2025 Re vised Dec 4, 2025 Accepted Jan 30, 2026 K eyw ords: Cate gorical boosting Extreme gradient boosting Machine learning Quantum boosting Quantum boosting plus Quantum machine learning V oting classier ABSTRA CT This study in v estig ates ensemble learning using classical and quantum-inspired models to predict ight delays at T ribhuv an International Airport (TIA), Nepal. It combines traditional machine learning algorithms with quantum-based ap- proaches, quantum boosting (QBoost) and the h ybrid QBoostPlus, le v eraging quantum properties for f aster computation. The dataset includes ight records from 2020 to 2024 and Meteorological Aerodrome Reports (MET AR), analyzed across four sea- sons to capture delay patterns in domestic and international ights. A combined seasonal dataset assesses model generalization. Six mod- els; V otingCla ssier , adapti v e boosting (AdaBoost), xtreme gradient boosting (XGBoost), cate gorical boosting (CatBoost), QBoost, and QB oostPlus are e v al- uated based on accurac y , precision, recall, F1 score, area under the curv e(A UC), and e x ecution time. CatBoost achie v ed high accurac y (up to 0.97) b ut slo wer e x ecution (up to 10,570.63 ms). QBoostPlus pro vides competiti v e A UC scores (0.83–0.95) with f aster e x ecution, impro ving speed by up to 99.94% and gen- erating predictions in as little as 6.46 ms. Al though quantum-inspired models ha v e slightly lo wer accurac y , their computational ef cienc y and stability sho w strong potential for real-time ight delay prediction. This is the rst study ap- plying quantum-inspired ensemble learning to Nepalese a viation data, sho wing promise for re gional airports with limited infrastructure. This is an open access article under the CC BY -SA license . Corresponding A uthor: Nanda Bikram Adhikari Department of Electronics and Computer Engineering, IOE, Pulcho wk Campus, T ribhuv an Uni v ersity Lalitpur 44600, Nepal Email: adhikari@ioe.edu.np 1. INTR ODUCTION T ribhuv an International Airport (TIA) in Kathmandu, Nepal, serv es as the nation’ s primary interna- tional g ate w ay , connecting to o v er 40 global destinations. Despite its strate gic role, TIA f aces operational challenges due to a single sloped runw ay , absence of an instrument landing system (ILS), and increasing traf- c demand. According to of cial TIA data, international passenger traf c gre w by 9.29% in 2024, a v eraging 13,598 passengers per day [1]. This sur ge has intensied congestion, delays, and resource limitations, empha- sizing the need for intelligent ight delay prediction systems to support ef cient airport operations. Flight delay prediction has been e xte nsi v ely studied using v arious machine learning (ML) t echniques. Deep learning approaches, such as con v olutional neural netw ork–long short-term memory (CNN-LSTM) frame- w orks, ha v e sho wn promi sing results in forecasting delays based on historical data [2], [3]. Hybrid ML models J ournal homepage: https://telk omnika.uad.ac.id/inde x.php/TELK OMNIKA Evaluation Warning : The document was created with Spire.PDF for Python.
528 ❒ ISSN: 1693-6930 combining dif ferent algorithms further impro v e prediction accurac y [4], [5], while ensemble learning methods lik e gradient boosting and incremental learning ef fecti v ely capture comple x delay patterns [6], [7]. Addition- ally , studies le v eraging a viation big data ha v e enhanced delay prediction models [8], and in v estig ations into the impact of short-term features ha v e rened model performance [9], [10]. Flight trajectory prediction has also beneted from h ybrid deep learning techniques, impro ving four -dimensional (4D) trajectory forecasts [11], and spatiotemporal propag ation learning has been proposed for netw ork-wide delay prediction [12]. Recent adv ancements include transformer architectures for temporal modeling in airport delay prediction [13], [14]. In parallel, quantum machine learning (QML) techniques are emer ging as no v el approaches for aerodynamic classication and time series forecasting in a viation. Quantum support v ector machines (QSVM) and data re-uploading quantum methods ha v e demonstrated potential in handling lar ge-scale spatiotemporal data and traf c forecasting [15], [16], opening ne w a v enues for ight delay modeling. Compared to h ybrid models lik e stacking and bagging [17], [18], quantum boosting plus (QBoost- Plus) int e grat es quantum-inspired optimization with ensemble fusion, using area under the curv e (A UC)-based weighting to impro v e speed and accurac y without iterat i v e retraining [19]. T ransformer -based ensembles [13], [20] ha v e sho wn high accurac y in ight delay prediction b ut with hea vy computational costs, limiting real-time use in constrained en vironments. Recent QML de v elopments [16] in transportation and time-series forecasting, such as quantum data re-uploading, of fer competiti v e accurac y and f aster con v er gence o v er classical models. Hybrid quantum models lik e quantum k ernel l ong short-term memory (QK-LSTM) ha v e impro v ed predicti v e ef cienc y and reduced computational costs in climate time-series tasks [21]. Quantum long short-term mem- ory (QLSTM) sho ws f aster con v er g e nce and lo wer test loss than classical LSTM on solar forecasting [22], while quantum sequential recurrent neural netw ork (QSe gRNN) achie v es comparable or better accurac y with fe wer parameters [23]. These results highlight QML ’ s potential to o v ercome late n c y and scalability issues in transportation and a viation forecasting. This study presents the rst application of quantum-inspired ensemble learning for ight delay predic- tion in Nepal, focusing on TIA. Existing ML models often lack the speed and scalability needed for real-time use in resource-constrained settings. T o address this, we propose QBoostPlus a h ybrid frame w ork combin- ing classical ensembles with quantum-inspired optimization to reduce comple xity . Using a multi-season ight and meteorological aerodrome reports (MET AR) weather datasets, the model impro v es both accurac y and ef- cienc y in delay forecasting. It supports real-tim e decision-making and is adaptable to other re gional airports, adv ancing smart airport initiati v es. The k e y contrib utions of this study include: (i) inte grating classical ensemble models with quantum- inspired optimization for delay prediction; (ii) proposing QBoostPlus for f ast and accurate delay prediction suitable for real-tim e use; (iii) e v aluating seasonal and aggre g ate datasets to assess model generalization; and (i v) demonstrating trade-of fs between accurac y and e x ecution time to inform h ybrid deplo yment strate gies. 2. METHOD 2.1. Dataset and pr epr ocessing This study utilized tw o primary datasets: the A viBit T raf c Solutions Dataset, which includes 12 ight-related features such as ight number , date, scheduled departure and arri v al times, tra v el time, origin and destination, distance and actual arri v al time in the training set, and a test set with the same features e xcept actual arri v al time. The second is the MET AR dataset, containing 13 meteorological features from the TIA MET AR station, including visibility , sk y conditions, tempera ture, wind, pressure, humidity , and precipitation. Both datasets were clean, with no missing v alues or duplicates. Data preprocessing in v olv ed mer ging the datasets into a single data frame (DataFrame), synchronizing weather data to coordinated uni v ersal time (UTC) and rounding timestamps to the nearest hour . More than 200,000 communication records were collected from 2020 to 2024. K e y subsets that signicantly contrib ute to the model’ s performance include: seasonal data with 9,522 training and 3,742 test samples, and a combined approach with 3,978 training and 1,610 test samples. Feature engineering included encoding sk y conditions , imputing zero v alues, remo ving redundant features, and aligning weather stations with origin and destinati on airports. Feature scaling w as performed using the standard scaler (StandardScaler) to normalize input v ariables. F or feature selection, columns with e xcessi v e missing data were remo v ed, and the top 14 features were selected based on mutual information (MI) scores. MI measures the de gree of dependenc y between each fe ature and the tar get v ariable, allo wing us to prioritize inputs that contrib ute most to predicting delays. This approach impro v es model interpretability by TELK OMNIKA T elecommun Comput El Control, V ol. 24, No. 2, April 2026: 527–535 Evaluation Warning : The document was created with Spire.PDF for Python.
TELK OMNIKA T elecommun Comput El Control ❒ 529 identifying features with the strongest predicti v e relationships, of fering insights into which ight and weather v ariables most inuence delays. The nal selected features comprised one ight characteristic distance; three origin weather features de w point temperature, precipitation, and fe w clouds at le v el 1; and ten destination weather features dry b ulb temperature, de w point temperature, wind speed, wind direction, wind gust, pressure, visibility , precipitation, relati v e humidity , and scattered clouds at le v el 1. W e ackno wledge that seasonal imbalance in the dataset (e.g., higher ight v olumes in spring and summer compared to winter and autumn) may inuence MI scoring, as features dominant in peak seasons could be o v eremphasized. T o mitig ate this, feature selection w as performed on both se asonal subsets and the combined dataset to ensure generalization across v arying traf c conditions. Figure 1 illustrates the architecture of the QBoostPlus fram e w ork, which inte grates quantum-ins pired optimization within a lightweight ensemble model to enhance con v er gence, generalization, and operational ef cienc y . Figure 1. System o w diagram 2.2. Model b uilding and implementation T o comprehensi v ely e v al uate predicti v e performance, we implemented three types of models: classical ML models, the QBoost model, and h ybrid approaches. 2.2.1. Classical models In our study , we utilized classical ensemble models including adapti v e boosting (AdaBoost), e xtr eme gradient boosting (XGBoost), cate gorical boosting (CatBoost), and v oting classier (V otingClassier) for clas- sication. AdaBoost sequentially impro v ed performance by focusing on misclassied instances [24], [25]. XGBoost o f fered high accurac y and ef cienc y through gradient boosting with re gularization [24], [26]. Cat- Boost ef fecti v ely handled cate gorical features using ordered boosting [26]. The V otingClassier combined predictions from multiple models using hard or soft v oting, enhancing o v erall stability and accurac y [24], [25]. 2.2.2. QBoost model QBoost is a quantum-inspired classication algorithm that reformulates problems into quadratic un- constrained binary optimization (Q UBO) format for quantum annealing on quantum processing units (QPUs). Hybrid classical–quantum ensemble learning for r eal-time ight delay pr ediction at ... (P avan Khanal) Evaluation Warning : The document was created with Spire.PDF for Python.
530 ❒ ISSN: 1693-6930 Due to limited access to D-W a v e hardw are, we used the simulated annealing sampler (SimulatedAnneal- ingSampler) from the dimod library , which emulates quantum annealing on classical hardw are while preserv- ing the Q UBO frame w ork [27]. Although it mimics quantum concepts lik e superposition and entanglement, it lacks true quantum features such as tunneling and lar ge-scale parallelism, limiting scalability . Ne v ertheless, this approach allo ws ef fecti v e testing of quantum-inspired models for classication and optimization. 2.2.3. Hybrid model – QBoostPlus QBoostPlus is a h ybrid ensemble classication model that combines multiple weak classiers using A UC-based weighting. Instead of relying on a single best model, it e v aluates each classier’ s A UC on a v ali- dation set and assigns weights through e xponential scaling, gi ving more inuence to stronger classiers. This weighting strate gy aligns with ensemble fusion theory , where model contrib utions are often scaled by perfor - mance m etrics to maximize o v erall predicti v e po wer [28], [29]. Predictions are generated by aggre g ating the weighted outputs, enhancing both di v ersity and accurac y . Unlik e traditional boosting, QBoostPlus a v oids it- erati v e training and instead focuses on performance-dri v en fusion of pre-trained models. The implementation in v olv es selecting the best classier based on A UC, optionally adding anot her , and e v aluating the model’ s per - formance and e x ecution time. F ormal equation of QBoostPlus: ˆ y ( x ) = sign   N X i =1 w i · h i ( x ) ! (1) where, N = number of weak classiers, h i ( x ) = prediction (or decision function output) of the i th classier on input x , w i = weight assigned to the i th classier bas ed on its A UC score (normalized so P i w i = 1 ), and ˆ y ( x ) = nal predicted label (e.g., +1 or − 1 ). Probability estimation (using a sigmoid function with temperature scaling): P ( y = 1 | x ) = 1 1 + e x p − 1 T P N i =1 w i · h i ( x ) (2) where, T = temperature parameter controlling the softness of probabilities. 2.2.4. Ev aluation metrics All models were e v aluated using standard classication metrics, including accurac y , precision, recall, F1-score, and A UC-recei v er operating characteristic (R OC), to assess their predicti v e performance compre- hensi v ely . In addition to these e v aluation metrics, e x ecution time w as recorded to compare the computational ef cienc y of classical, quantum, and h ybrid models, pro viding insights into both ef fecti v eness and practicality for real-w orld applications. 2.3. T oolset and system conguration The en vironment used V isual Studio Code (v1.95.3), Python 3.x, and libraries such as NumPy , pan- das, scikit-learn, and Simulated Annealing from the dimod library . Experiments were run on a system with an Intel Core i5-1035G1 central processing unit (CPU) (1.00 GHz, up to 1.19 GHz), 8 GB random access memory (RAM), and W indo ws 11, which supported both ML and quantum-inspired simulations ef ciently . 3. RESUL T AND DISCUSSION 3.1. Analysis of combined appr oach f or all seasons The combined approach inte grates ight data from all seasons into a single training and testing frame- w ork, enabling a holistic assessment of delay patterns. By aggre g ating seasonal v ariations, this approach captures recurring operational characteristics such as airport congestion and systemic inef ciencies while ben- eting from a lar ger and more di v ers e dataset. As a result, the models e xhibit impro v ed stability and reduced susceptibility to o v ertting. In addition, emplo ying a single unied model simplies deplo yment and lo wers computational o v erhead, which is essential for real-time operational use. TELK OMNIKA T elecommun Comput El Control, V ol. 24, No. 2, April 2026: 527–535 Evaluation Warning : The document was created with Spire.PDF for Python.
TELK OMNIKA T elecommun Comput El Control ❒ 531 Figure 2 illustrates t h e A UC–R OC perf o r mance of the e v aluated m od e ls under the combined set ting. XGBoost, CatBoost, the V otingClassier , and QBoostPlus demonstrate the strongest discriminati v e capability , indicating reliable separation between delayed and on-time ights. In contrast, AdaBoost and QBoost sho w comparati v ely weak er performance, suggesting limited rob ustness under aggre g ated seasonal conditions. Figure 3 present s the relationship between predicti v e performance and e x ecution time. QBoostP lus achie v es the f astest inference time, substantially outperforming other ensemble models. Although CatBoost and the V otingClassier attain comparable predicti v e accurac y , their signicantly higher e x ecution times limit their suitability for latenc y-sensiti v e en vironments. These result s indicate that QBoostPlus pro vides an ef fecti v e balance between predicti v e capabilit y and computational ef cienc y , making it a strong candidate for real-time ight delay prediction. Figure 2. A UC R OC curv e of combined approach Figure 3. Classication performance vs. e x ecution time of combined approach Hybrid classical–quantum ensemble learning for r eal-time ight delay pr ediction at ... (P avan Khanal) Evaluation Warning : The document was created with Spire.PDF for Python.
532 ❒ ISSN: 1693-6930 3.2. Enhancing statistical r ob ustness of model e v aluation T o ensure reliable performance estimation, cross-v alidation and statistical signicance testing were emplo yed. T able 1, sum marizes the a v erage accurac y and standard de viation obtained from 5-fold and 10-fold cross-v alidation, along with paired t-test results. The models demonstrate consistent generalization, with mean cross-v alidation accuracies ranging from approximately 85% to 93%. QBoostPlus achie v es the highest a v erage accurac y across both v alidation settings, while lo w standard de viations indicate stable performance. P aired t-test results sho w no signicant dif ferences between 5-fold and 10-fold v alidation (all p -v alues > 0 . 05 ), conrming the reli ability of the re- ported estimates. These results highlight that QBoostPlus deli v ers strong predicti v e performance with ef cient computational cost, supporting its suitability for practical deplo yment. T able 1. Model performance with cross-v alidation and signicance testing Model 5-fold a v erage 5-fold standard de viation 10-fold a v erage 10-fold standard de viation t-T est v alue Signicance le v el (p) AdaBoost 0.8283 0.0045 0.8313 0.0110 -0.7045 0.4936 CatBoost 0.8585 0.0085 0.8610 0.0121 -0.4274 0.6776 XGBoost 0.8522 0.0120 0.8532 0.0146 -0.1292 0.9000 V otingClassier 0.8595 0.0113 0.8612 0.0131 -0.2448 0.8123 QBoost 0.8512 0.0129 0.8668 0.0103 -2.1343 0.0741 QBoostPlus 0.9168 0.0197 0.9301 0.0242 -1.0472 0.3215 3.3. Seperate analysis of each season Flight delays at TIA are strongly inuenced by seasonal f act ors. W inter fog, spring storms, summer congestion and heat, and autumnal weather transi tions introduce distinct operational challenges. T o account for these ef fects, a season-wise e v aluation w as conducted to assess conte xt-specic model beha vior . As sho wn in T able 2, reports the classication performance and e x ecution time of each model across the four seasons. QBoostPlus consistently demonstrates strong predicti v e performance, achie ving its highest accurac y during the summer season while maintaining competiti v e results in winter , spring, and autumn. Im- portantly , it preserv es e xceptionally lo w e x ecution times across all seasonal datasets, highlighting its rob ustness under v arying operational conditions. T able 2. Classication performance and e x ecution time of models across dif ferent seasons Season Model A UC Accurac y F1-score Precision Recall Ex ecution time (ms) W inter AdaBoost 0.88 0.89 0.76 0.81 0.73 495.72 CatBoost 0.90 0.91 0.81 0.88 0.88 4660.70 XGBoost 0.89 0.92 0.84 0.88 0.81 219.86 V otingClassier 0.90 0.92 0.83 0.89 0.79 4636.98 QBoost 0.89 0.90 0.81 0.83 0.79 55.75 QBoostPlus 0.90 0.91 0.82 0.87 0.79 11.20 Spring AdaBoost 0.82 0.90 0.63 0.66 0.62 360.62 CatBoost 0.89 0.92 0.68 0.74 0.65 4184.21 XGBoost 0.88 0.92 0.69 0.73 0.67 208.22 V otingClassier 0.89 0.92 0.69 0.75 0.65 4783.87 QBoost 0.88 0.90 0.62 0.66 0.60 53.28 QBoostPlus 0.89 0.91 0.63 0.68 0.61 10.59 Summer AdaBoost 0.94 0.96 0.81 0.88 0.77 570.90 CatBoost 0.93 0.97 0.85 0.98 0.78 4091.32 XGBoost 0.95 0.96 0.82 0.86 0.79 125.19 V otingClassier 0.95 0.97 0.83 0.94 0.77 4466.01 QBoost 0.95 0.95 0.82 0.81 0.84 58.26 QBoostPlus 0.95 0.97 0.86 0.88 0.84 8.45 Autumn AdaBoost 0.76 0.87 0.64 0.69 0.62 274.94 CatBoost 0.83 0.89 0.70 0.77 0.66 3537.77 XGBoost 0.82 0.89 0.70 0.74 0.68 162.01 V otingClassier 0.83 0.89 0.71 0.76 0.68 4010.41 QBoost 0.82 0.87 0.68 0.70 0.68 55.21 QBoostPlus 0.83 0.88 0.69 0.71 0.68 9.09 TELK OMNIKA T elecommun Comput El Control, V ol. 24, No. 2, April 2026: 527–535 Evaluation Warning : The document was created with Spire.PDF for Python.
TELK OMNIKA T elecommun Comput El Control ❒ 533 While CatBoost and the v oting classier occasionally achie v e comparable accurac y , their substantially higher inference times re d uc e their practicality in en vironments with limited computational resources and strict real-time constraints [13]. Classical ensemble methods such as AdaBoost sho w higher precision in certain seasons b ut suf fer from reduced recall, particularly during autumn, indicating sensiti vity to class imbalance and temporal v ariabilit y [6], [24]. QBoostPlus maintains a balanced trade-of f between precis ion and recall, resulting in stable F1-scores e v en in challenging seasonal conditions. This beha vior aligns with prior studies reporting the dif culty of delay prediction under imbalanced and temporally heterogeneous data distrib utions [6], [24]. Compared with recently proposed transformer -based approaches [13], which of fer strong predicti v e performance at the e xpense of high computational comple xity , QBoostPlus deli v ers comparable accurac y with signicantly lo wer latenc y . Ov erall, the seasonal analysis conrms that QBoostPlus ef fecti v ely adapts to di v erse operational con- te xts while preserving computational ef cienc y . Its application to ight delay prediction at TIA represents, to the best of our kno wledge, the rst use of a quantum-inspired ensemble learning approach in the Nepalese a vi- ation domain. These res u l ts suggest strong potential for broader adoption in infrastructure-constrained airports where scalability and real-time responsi v eness are critical [30]. 4. CONCLUSION This study e xplored classical and quantum-inspired ML models for ight delay prediction at TIA, introducing a h ybrid frame w ork that balances computational ef cienc y with predicti v e accurac y . The ndings highlight the potential of quantum-inspired approaches for time-sensiti v e a viation tasks, particularly in airports with limited resources. This w ork contrib utes to adv ancing intelligent, adapti v e delay prediction systems tai- lored to comple x airport operations. Future research should focus on implementing this frame w ork using actual quantum hardw are and e xtending it to other re gional airports to enhance scalability and practical utility . A CKNO WLEDGMENTS The authors thank T ribhuv an International Airport (TIA) and the Department of Hydrology and Me- teorology (DHM), Nepal, for pro viding communication and meteorological aerodrome report (MET AR) data for this study . FUNDING INFORMA TION Authors state no funding in v olv ed. A UTHOR CONTRIB UTIONS ST A TEMENT This journal uses the Contri b ut or Roles T axonomy (CRediT) to recognize indi vidual author contrib u- tions, reduce authorship disputes, and f acilitate collaboration. Name of A uthor C M So V a F o I R D O E V i Su P Fu P a v an Khanal ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Nanda Bikram Adhikari ✓ ✓ ✓ ✓ ✓ ✓ C : C onceptualization I : I n v estig ation V i : V i sualization M : M ethodology R : R esources Su : Su pervision So : So ftw are D : D ata Curation P : P roject Administration V a : V a lidation O : Writing - O riginal Draft Fu : Fu nding Acquisition F o : F o rmal Analysis E : Writing - Re vie w & E diting CONFLICT OF INTEREST ST A TEMENT Authors state no conict of interest. Hybrid classical–quantum ensemble learning for r eal-time ight delay pr ediction at ... (P avan Khanal) Evaluation Warning : The document was created with Spire.PDF for Python.
534 ❒ ISSN: 1693-6930 D A T A A V AILABILITY The data used in this st udy were obta ined from T ribhuv an International Airport (TIA). Due to polic y restrictions, t he dataset is not publicly a v ailable b ut may be pro vided upon reasonable re qu e st to the corre- sponding author , subject to institutional or re gulatory appro v al. REFERENCES [1] S. Prasain, “Kathmandu airport nears 5 million international yers, ” The Kathmandu P ost . Feb . 13, 2025. [Online]. A v ailable: https://kathmandupost.com/mone y/2025/02/13/kathmandu-airport-nears-5-million-international-yers (Accessed: May 24, 2025). [2] A. A yaydın and M. A. Akcayol, “Deep Learning Based F orecasting of Delay on Flights, ” Bilis ¸ im T eknolojileri Der gisi , v ol. 15, no. 3, pp. 239–249, 2022, doi: 10.17671/g azibtd.1060646. [3] Q. Li, X. Guan, and J. Liu, “ A CNN-LSTM frame w ork for ight delay prediction, ” Expert Systems with Applications , v ol. 227, p. 120287, 2023, doi: 10.1016/j.esw a.2023.120287. [4] R . K. Jha, S. B. Jha, V . P ande y , and R. F . Babiceanu, “Flight Delay Prediction using Hybrid Machine Learning Approach: A Case Study of Major Airlines in the United States, ” 2024, arXiv , doi: 10.48550/ARXIV .2409.00607. [5] D. B. Bisandu and I. Moulitsas, “ A Hybrid Ensemble Machine Learning Approach for Arri v al Flight Delay Cl assication Prediction Using V oting Aggre g ation T echnique, ” in AIAA A VIA TION 2023 F orum. American Institute of Aer onautics and Astr onautic , Jun. 2023, doi: 10.2514/6.2023-4326. [6] R. Khan, S. Akbar , and T . A. Zahed, “Flight Delay Prediction Based on Gradient Boosting Ensemble T ech- niques, ” 2022 16th Internat ional Confer ence on Open Sour ce Systems and T ec hnolo gies (ICOSST) , 2022, pp. 1-5, doi: 10.1109/ICOSST57195.2022.10016828. [7] D. W ang, M. W ang, X. W ang, and P . Y ang, “Ensemble of Incremental Learning Algorithm for Flight Delay Prediction, ” Beijing Gongye Daxue Xuebao/J ournal of Beijing Univer sity of T ec hnolo gy , v ol. 46, no. 11, pp. 1239–1245, 2020, doi: 10.11936/bjutxb2019030009. [8] N. K umar , Hanumanth, Nag areddy , and J. P atil, “Flight Delay Prediction Based On A viation Big Data And Machine Learning, ” J ournal of Scientic Resear c h and T ec hnolo gy , v ol. 1, no. 7, pp. 58–67, 2023. [9] D. Sch ¨ osser and J. Sch ¨ onber ger , “On the Performance of Machine Learning Based Flight Delay Prediction – In v estig ating the Impact of Short-T erm Features, ” Pr omet - T r af c and T r ansportation , v ol. 34, no. 6, pp. 825–838, 2022, doi: 10.7307/ptt.v34i6.4132. [10] E. Esmaeilzadeh and S. Mokhtarimousa vi, “Machine Learning Approach for Flight Departure Delay Prediction and Analysis, ” T r ansportation Resear c h Recor d: J ournal of the T r ansportation Res ear c h Boar d , v ol. 2674, no. 8, pp. 145–159, 2020, doi: 10.1177/0361198120930014. [11] H. Shaen ya and A. C. Re g an, “4D ight trajectory prediction using a h ybrid Deep Learning prediction method based on ADS-B technology: A case study of Hartseld–Jackson Atlanta International Airport (A TL), ” T r ansportation Resear c h P art C: Emer ging T ec hnolo gies , v ol. 144, p. 103878, No v . 2022, doi: 10.1016/j.trc.2022.103878. [12] Y . W u, H. Y ang, Y . Lin, and H. Liu, “Spatiotemporal Propag ation Learning for Netw ork-W ide Flight Delay Prediction, ” IEEE T r ansactions on Knowledg e and Data Engineering , v ol. 36, no. 1, pp. 386–400, 2024, doi: 10.1109/TKDE.2023.3286690. [13] K. Liu et al. , “ Airport Delay Prediction with T emporal Fusion T ransformers, ” in Pr oceedings of the 17th A CM SIGSP A TIAL In- ternational W orkshop on Computational T r ansportation Science GenAI and Smart Mobility Session , Oct. 2024, pp. 5–11, doi: 10.1145/3681772.3698212. [14] L. W ang, A. T ien, and J. Chou, “Multi-Airport Delay Prediction with T ransformers, ” AIAA A VIA TION 2022 F orum. American Institute of Aer onautics and Astr onautics , 2022, doi: 10.2514/6.2022-3707. [15] X. J. Y uan et al. , “Quantum Support V ector Machines for Aerodynamic Classication, ” Intellig ent Computing , v ol. 2, 2023, doi: 10.34133/icomputing.0057. [16] N. Schetakis et al. , “Data re-uploa ding in Quantum Machine Learning for time series: application to traf c forecasting, ” 2025, arXiv , doi: 10.48550/ARXIV .2501.12776. [17] P . Schw arz, “Stacking ensemble classication applied to us ight delay prediction during the co vid-19 pandemic, ” M.Sc. thesis, De- partment of Statistics, F aculty of Socia l Sciences, [Sweden], Spring 2022. [Online]. A v ailable: https://www.diva- portal. org/smash/get/diva2%3A1715076/FULLTEXT01.pdf (Accessed: May 24, 2025). [18] X. Cai, Q. Jin, and W . Zhang, “T raf c o w prediction: A method using bagging-based ensemble learning model, ” Science J ournal of Applied Mathematics and Statistics , v ol. 12, no. 5, pp. 72–79, October 2024, doi: 10.11648/j.sjams.20241205.11. [19] S. K. Jha and N. B. Adhikari, “ Analysis of Ensembled Learning using Classical and Quantum Boosting Approaches for Diabetes Mellitus Prediction, ” Resear c h Squar e Platform LLC , Apr . 05, 2024, doi: 10.21203/rs.3.rs-4199185/v1. [20] Q. Zhong, Y . Y u, Y . Huang et al., “Prediction and optimization of ci vil a viation ight delays based on machine learning algorithms, ” International J ournal of Computational Intellig ence Systems , v ol. 18, p. 189, 2025, doi: 10.1007/s44196-025-00932-2 [21] Y . C. Hsu, N. Y . Chen, T . Y . Li, P . H. H. Lee, and K. C. Chen, “Quantum K ernel-Based Long Short-term Memory for Climate T ime- Series F orecasting, ” Pr oceedings - 2025 International Confer ence on Quantum Communications, Networking , and Computing , QCNC 2025 , 2025, pp. 421–426, doi: 10.1109/QCNC64685.2025.00072. [22] S. Z. Khan et al. , “Quantum long short-term memory (QLSTM) vs. classical LSTM in time series forecasting: a comparati v e study in solar po wer forecasting, ” F r ontier s in Physics , v ol. 12, 2024, doi: 10.3389/fph y .2024.1439180. [23] K.-H. Moon, S.-G. Jeong, and W .-J. Hw ang, “QSe gRNN: quantum se gment recurrent neural netw ork for time series forecasting, ” EPJ Quantum T ec hnolo gy , v ol. 12, no. 1, 2025, doi: 10.1140/epjqt/s40507-025-00333-6. [24] R. Natras, B. Soja, and M. Schmidt, “Ensemble Machine Learning of R andom F orest, AdaBoost and XGBoost for V ertical T otal Electron Content F orecasting, ” Remote Sensing , v ol. 14, no. 15, p. 3547, Jul. 2022, doi: 10.3390/rs14153547. [25] S. Ghosh, M. K. Gourisaria, B. Sahoo, and H. Das, “ A pragmatic ensemble learning approach for rainf all prediction, ” Disco ver Internet of Things , v ol. 3, no. 1, 2023, doi: 10.1007/s43926-023-00044-3. [26] M. Mamdouh, M. Ezzat, and H. A.Hefn y , “ A no v el intelligent approach for ight delay prediction, ” J ournal of Big Data , v ol. 10, no. 1, Dec. 2023, doi: 10.1186/s40537-023-00854-w . TELK OMNIKA T elecommun Comput El Control, V ol. 24, No. 2, April 2026: 527–535 Evaluation Warning : The document was created with Spire.PDF for Python.
TELK OMNIKA T elecommun Comput El Control ❒ 535 [27] D-W a v e Systems Inc., “Ocean SDK: Simul ated annealing sampler reference, ” D-W ave Documentation . [Online]. A v ailable: https://docs.dw a v equantum.com/en/latest/inde x.html (Accessed: May 24, 2025). [28] J. Zhao, J. Jin, S. Chen, R. Zhang, B. Y u, and Q. Liu, “ A weighted h ybrid ensemble method for classifying imbalanced data, ” Knowledg e-Based Systems , v ol. 203, p. 106087, Sep. 2020, doi: 10.1016/j.knosys.2020.106087. [29] X. W ang, Y . Li, and J. Zhang, “W eighted ensemble based on dif ferentiated sampling rates for imbalanced classication a nd appli- cation to credit risk assessment, ” Expert Systems with Applications , v ol. 262, 2025, doi: 10.1016/j.esw a.2024.125595. [30] X. Y . Zhang and M. M. W ang, “ An ef cient combination st rate gy for h ybrid quantum ensemble classier , ” International J ournal of Quantum Information , v ol. 21, no. 6, 2023, doi: 10.1142/S0219749923500272. BIOGRAPHIES OF A UTHORS P a v an Khanal recei v ed his Bachelor’ s de gree in Computer Engineering from T ribhuv an Uni v ersity , Nepal, in 2009, and his Master’ s de gree in Computer Syst ems and Kno wledge Engi- neering from the Institute of Engineering, Pulcho wk Campus, T ribhuv an Uni v ersity , in 2025. He is currently serving as a senior IT of cer at the Ci vil A viation Authority of Nepal. His current research interests include computer netw orking, c ybersecurity , machine learning, and quantum computing. He can be contacted at email: pkhanal2008@gmail.com. Nanda Bikram Adhikari recei v ed an M.Sc. in Engineering de gree from the State Engi- neering Uni v ersity of Armenia in 1994 and a Ph.D. from Nago ya Uni v ersity , Japan, in 2004. From 2004 to 2007, he w ork ed as a res earch fello w at the National Institute of Information and Com- munications T echnology , T ok yo, Japan, de v eloping a rainf all rate retrie v al algorithm for the global precipitation measurement satellite mission. Since 2007, he has been an Associate Professor in the Department of Electronics and Computer Engineering, Institut e of Engineering, Pulcho wk Campus, T ribhuv an Uni v ersity . He has published o v er 40 articles in high-impact SJR-rank ed ISI and Sco- pus journals and IEEE conferences, including Ra dio Science, IEEE T ransactions on Geoscience and Remote Sensing, A GU, and MDPI Electronics. He also serv es as a re vie we r for more than 20 presti- gious journals, including IEEE, Else vier , and Springer . He has authored four books: Outlook of Re- mote Sensing, Fundamentals of Micro w a v e Engi neering, Performance Analysis of Cogniti v e Radio Netw orks for Resource Sharing, and Design and Implementat ion of Multi-channel Acti v e Electrode EEG De vice. His research interests include 5G and be yond netw orks, micro w a v e wireless com- munications, softw are-dened netw orks, IoT applications, quantum computing, and AI-based signal processing. He can be contacted at email: adhikari@ioe.edu.np. Hybrid classical–quantum ensemble learning for r eal-time ight delay pr ediction at ... (P avan Khanal) Evaluation Warning : The document was created with Spire.PDF for Python.