Customer Purchase Prediction Using Optimized Decision Tree with Particle Swarm Optimization

This study examines the use of Particle Swarm Optimization (PSO) to improve the accuracy of Decision Tree classifiers in analyzing consumer purchase in a supermarket. The study is aimed at predicting the sales of a well-known cooking oil brand and attempts to optimize hyperparameters in order increase the performance of the model. The dataset, consisting of sales data from a supermarket loyalty program in January 2022, was analyzed. PSO was utilized to optimize hyperparameters, including maximum tree depth and minimum sample split, hence improving the model's predictive performance. The optimized model showed a significant enhancement, as the accuracy improved from 70.67% to 82.15%. The examination of the confusion matrix and classification report showed that the model has a cautious inclination towards making predictions, exhibiting a high level of accuracy in predicting non-purchases while demonstrating less dependability in properly detecting purchases. The results highlight the capacity of PSO to enhance machine learning models for retail analytics, providing vital insights for strategic decision-making in marketing and sales. This research has significant implications for the creation of focused marketing efforts and the implementation of inventory management tactics, especially during periods of low sales, such as the months after the holiday season. The results emphasize the importance of utilizing sophisticated analytical methods to improve comprehension of consumer behavior and stimulate retail achievement.