EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Dual‐Timescale Behaviour Modelling for Customer Conversion Prediction Using Hybrid Deep Learning Models in Loyalty and Subscription‐Based Systems

Mehwish Iqra Taha, Shahbaz Pervez, Muneer Ahmad, Seyed Ebrahim Hosseini, Mohsin Iftikhar, Peer Azmat Shah

Complexity · 2026-01 · 2026 (1)

DOI: 10.1155/cplx/7712351 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Complexity, online erschienen am 16.11.2026. Verfasst von Mehwish Iqra Taha und 5 weiteren Autor:innen. Die Arbeit nennt 41 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Digital platforms increasingly rely on artificial intelligence (AI) and machine learning (ML) to analyse behaviour data and predict conversion outcomes within loyalty and subscription‐based systems. Accurate customer conversion prediction has become critical as businesses increasingly rely on behavioural data generated through continuous digital interactions. Despite recent advancements in predictive analytics, existing research predominantly models customer behaviour at a single temporal scale, focusing either on short‐term session‐level interactions or long‐term engagement patterns. This limitation restricts the ability to capture the combined influence of immediate behavioural intent and accumulated customer engagement on conversion outcomes. To address these limitations, this study proposes a Dual‐Timescale hybrid deep learning framework that integrates short‐term session‐level behavioural features with long‐term cumulative behavioural engagement features. The methodology incorporates a dual‐input model architecture, comparative benchmarking against traditional ML models and the application of explainable AI techniques to interpret model predictions. Experimental evaluation using the Instacart online grocery shopping dataset, where conversion is operationalised as subsequent observed purchase behaviour, demonstrates that the proposed Dual‐Timescale modelling framework improves performance compared to both traditional ML baseline models and single‐timescale neural architectures. The Dual‐Timescale model achieved a ROC‐UC of 0.8750 on the holdout test, compared with 0.8413 for the strongest traditional baseline, Gradient Boosting. Under the 10‐fold GroupKFold validation, the model achieved a mean ROC‐AUC of 0.8710 ± 0.0034, representing a mean improvement of 0.0296 over Gradient Boosting ( p < 0.001). These findings demonstrate that integrating short‐term behavioural patterns with long‐term engagement features improves customer conversion prediction and provides a more comprehensive representation of repeated customer purchase behaviour, with potential applicability to retention‐oriented loyalty and subscription settings. Future research may extend this work by incorporating real‐time behavioural data streams, evaluating the framework across multiple industry domains and exploring adaptive or reinforcement learning mechanisms for dynamic customer engagement modelling. These directions position the proposed Dual‐Timescale framework as an approach for advancing predictive analysis in loyalty and subscription‐based digital platforms.

Zitieren

APA

Taha, M. I., Pervez, S., Ahmad, M., Hosseini, S. E., Iftikhar, M., & Shah, P. A. (2026). Dual‐Timescale Behaviour Modelling for Customer Conversion Prediction Using Hybrid Deep Learning Models in Loyalty and Subscription‐Based Systems. Complexity, 2026(1). https://doi.org/10.1155/cplx/7712351

Vancouver

Taha MI, Pervez S, Ahmad M, Hosseini SE, Iftikhar M, Shah PA. Dual‐Timescale Behaviour Modelling for Customer Conversion Prediction Using Hybrid Deep Learning Models in Loyalty and Subscription‐Based Systems. Complexity. 2026;2026(1). doi:10.1155/cplx/7712351

BibTeX
@article{Taha2026,
  author = {Taha, Mehwish Iqra and Pervez, Shahbaz and Ahmad, Muneer and Hosseini, Seyed Ebrahim and Iftikhar, Mohsin and Shah, Peer Azmat},
  title = {Dual‐Timescale Behaviour Modelling for Customer Conversion Prediction Using Hybrid Deep Learning Models in Loyalty and Subscription‐Based Systems},
  journal = {Complexity},
  year = {2026},
  volume = {2026},
  number = {1},
  doi = {10.1155/cplx/7712351}
}
Kostenlos starten →