Evaluating Call Duration Availability in Bank Marketing Prediction

Author

Arjun Ananthakrishnan, S Sreya and Dr. G. Paavai Anand, Dr. G. Paavai Anand

Abstract

 Call duration can improve retrospective bank marketing prediction while being unavailable when a customer is selected for a call. This study quantifies that distinction through a matched comparison of four classification pipelines on the original Bank Marketing dataset. Logistic Regression, Decision Tree, Random Forest and K Nearest Neighbors were trained with all 16 predictors and with duration removed. The 45,211 real records were divided into 28,934 fitting, 7,234 validation and 9,043 test records using shared stratified indices. Preprocessing was fitted within each training pipeline, and the without-duration candidate was selected by validation F1. Random Forest achieved the highest candidate F1 in both conditions, decreasing from 0.5862 with duration to 0.4569 without it; ROC AUC decreased from 0.9240 to 0.7971. After refitting the selected without-duration forest on 36,168 development records, test F1 was 0.4549, ROC AUC was 0.8001 and average precision was 0.4470, with precision 0.3725 and recall 0.5841. The always-no baseline achieved higher accuracy, 0.8830, but zero subscriber recall. The results show how feature availability and evaluation criteria alter the interpretation of an otherwise conventional benchmark. They support a reproducible applied baseline rather than a new algorithm or a fully validated pre-call system. Remaining last-contact fields, one random split and the absence of external validation constrain operational conclusions.

Keywords

bank marketing; feature availability; call duration; classification; class imbalance; average precision; random forest; reproducibility

Full Text:

Download Paper PDF

References

Breiman, L. (2001). Random forests. Machine Learning, 45, 5–32. https://doi.org/10.1023/A:1010933404324

Kapoor, S., & Narayanan, A. (2023). Leakage and the reproducibility crisis in machine-learning-based science. Patterns, 4(9), Article 100804. https://doi.org/10.1016/j.patter.2023.100804

Moro, S., Cortez, P., & Rita, P. (2014). A data-driven approach to predict the success of bank telemarketing. Decision Support Systems, 62, 22–31. https://doi.org/10.1016/j.dss.2014.03.001

Moro, S., Rita, P., & Cortez, P. (2014). Bank Marketing [Data set]. UCI Machine Learning Repository. https://doi.org/10.24432/C5K306

Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P., Weiss, R., Dubourg, V., Vanderplas, J., Passos, A., Cournapeau, D., Brucher, M., Perrot, M., & Duchesnay, É. (2011). Scikit-learn: Machine learning in Python. Journal of Machine Learning Research, 12, 2825–2830. https://jmlr.org/papers/v12/pedregosa11a.html

Saito, T., & Rehmsmeier, M. (2015). The precision-recall plot is more informative than the ROC plot when evaluating binary classifiers on imbalanced datasets. PLOS ONE, 10(3), Article e0118432. https://doi.org/10.1371/journal.pone.0118432

Scikit-learn developers. (n.d.-a). average_precision_score. Retrieved October 30, 2026, from https://scikit-learn.org/stable/modules/generated/sklearn.metrics.average_precision_score.html

Scikit-learn developers. (n.d.-b). RandomForestClassifier. Retrieved October 30, 2026, from https://scikit-learn.org/stable/modules/generated/sklearn.ensemble.RandomForestClassifier.html

Share your valuable work from Social Media Buttons