Predicting product popularity in e-commerce using machine learning models: A case study of Trendyol

With the rapid advancement of smart device technology, the daily use of the internet is increasingly becoming widespread each day. This growth in technology has completely transformed the investment landscape, directing many investors towards the e-commerce sector. Significant progress has been made in the e-commerce sector through these investments. This development involves covering the entire goods and services sector in e-commerce, continuously adding new products, and increasing the variety and quantity of products to provide services to users. In this vast sector composed of massive data, arranging products based on user needs and selecting the most suitable product for the user becomes challenging. To overcome this challenge, many e-commerce sites use product-based predictions. The popularity of a product, one of these predictions, is determined by the feedback from users who have previously used the product. Feedback influences the popularity of a product, determining its position in the display order shown to the user. In this study, data from 7200 products, consisting of 17 attributes related to the Trendyol e-commerce site, were used to investigate the impact of product features on the number of favorites. Regression models such as Bagging, Gradient Boosting Machines (GBM), Light Gradient Boosted Machine (LightGBM), and Extreme Gradient Boosting (XGBoost) were used to predict the number of favorites for products. Regression models were evaluated using metrics such as Mean Absolute Error (MAE), Mean Square Error (MSE), Root Mean Square Error (RMSE), and Mean Absolute Scaled Error (MASE). The analysis results showed that the models successfully predicted the number of favorites with a low error rate.Additionally, a new dataset was created by classifying the favorite numbers in the dataset, forming a dataset consisting of favorite number classes. This dataset was used to investigate the impact of favorite number classes on popularity. For this, classification models such as Decision Tree, GBM, LightGBM, Naive Bayes, Random Forest, SVM, and XGBoost were used. Models were evaluated using classification metrics including Accuracy, F1_score, and cross-validation score. The analysis results demonstrated that the models could predict the favorite number class of products with high accuracy.