Analysis of Purchasing Behavior of Corporate Members on Arabam.com

This study aims to analyze the package purchase and usage behaviours of 24,511 corporate members of Arabam.com, active between 2016 and 2024, to recommend the most suitable package type to users. To address data imbalance, SMOTE (Synthetic Minority Over-sampling Technique) was used, and class distribution was balanced. Various classification models were evaluated using the PyCaret library, and the XGBoost Classifier provided the best performance with an Area Under the Curve (AUC) of 0.8636. Although the CatBoost Classifier achieved a higher AUC of 0.8660 and superiority in other metrics, it was not preferred due to its long training time of 23.3280 seconds. The XGBoost model offers fast and optimized results with a training time of 1.29 seconds. In addition to recommending the most suitable package for corporate members, a new model was developed to predict users who will purchase vehicles through auctions. This model utilized features such as the number of favourited cars, the number of purchased provisions, the number of bids, and the number of purchased vehicles and generated new features through interactions among these features. Modelling was done using PyCaret, and the Random Forest Classifier provided the best results in terms of accuracy (98.86%), AUC (99.44%), F1 score , and other performance metrics. The analyses indicate that users' purchasing behaviours change with seasonal effects and have the potential to increase customer satisfaction. As a result of the study, it will be possible to recommend the most suitable package type for corporate members and predict whether corporate members will purchase cars, one of the most important revenue sources for Arabam.com. This research provides significant insights into Arabam.com's digital marketing strategies and customer relationship management.