A Voting-based Hybrid Machine Learning Model For Predicting Customer Purchasing Behavior

Articles in Press

Document Type : Research Paper

Authors

Department of Industrial Management, Faculty of Management, University of Tehran, Tehran, Iran.

10.22059/ier.2026.390688.1008186

Abstract

One of the major challenges of marketing and customer relationship management is to predict the customer's purchasing behavior. In this study, a hybrid model of machine learning is developed to predict customer purchasing behavior in a more accurate manner than traditional methods. To increase the accuracy of predictions, the proposed model combines several machine learning algorithms in the form of voting mechanisms, namely soft voting. Specifically, this model integrates Random Forest, Gradient Boosting, XGBoost, LightGBM, and CatBoost with different weights assigned to each model to increase the impact of stronger learners. Results indicate that this hybrid model outperforms single model algorithms, including Logistic Regression and Naive Bayes. Specifically, among all possible models, the soft voting model has achieved the best performance, with an Accuracy of 0.9483, a precision of 0.9547, a recall of 0.9247, and an F1 Score of 0.9392. Additionally, this study stresses the significant role of economic status and social impacts in the prediction of purchase behaviour. Furthermore, it should be noted that the role of digital platforms and social media in customers’ purchasing decisions, particularly in the digital age, should not be neglected. Understanding these complex factors, the hybrid model described in this paper is more accurate and reliable than traditional models. Results from this study can help marketers and retailers develop and implement their marketing strategy more adequately.

Keywords

Main Subjects

Iranian Economic Review

Articles in Press, Accepted Manuscript
Available Online from 06 January 2026

Files

Share

How to cite

Statistics