Prediction of Customer Purchasing Patterns for Retail Optimization Using Market Basket Techniques

This paper addresses the critical task of assessing customer purchasing behaviour in retail, which directly impacts marketing strategies and inventory management. The paper presents a data processing and analysis pipeline specifically designed for market basket analysis and association rule mining. Utilizing the Apriori algorithm, our system identifies frequent itemsets and association rules, revealing key insights into purchasing patterns. The pipeline employs AWS S3 for data storage, Apache Airflow for ETL processes, and SQLite3 for data analysis, with visualizations provided through Power BI dashboards. The proposed model effectively identifies complex product relationships and predicts customer behaviour, outperforming traditional methods. The integration of a Large Language Model (LLM) like GPT 3.5-Turbo for customer feedback processing enhances the insights derived. This study enhances marketing strategies, inventory management, and customer satisfaction by offering actionable recommendations for product bundling and cross-selling. The findings highlight the significant potential of advanced data processing and machine learning algorithms to transform retail analytics, providing businesses with reliable tools for informed decisionmaking. Notably, the strongest association rule identified was between Wine and Frozen Meals, as well as Canned Vegetables and Beer, with a Support score of 3.6%, Confidence of 50.7%, and Lift value of 3.04. Furthermore, it was observed that the accuracy of Customer Feedback Analysis using GPT 3.5-Turbo was 92% while the traditional method of using in-built python libraries(Naive Bayes Classifier) gave an accuracy of only 84%.