Customer purchase prediction and product recommendations

Wong, Ji Hin (2024) Customer purchase prediction and product recommendations. Final Year Project, UTAR.

Preview

PDF Download (3660Kb) | Preview

Abstract

Understanding customer behaviour is important for businesses that seeking growth and sustainability in today’s competitive market. This project has three main objectives where the first objective is to develop a customer purchase prediction model to estimate the probability of future purchases. The second objective is to conduct market basket analysis to gain insights into customer shopping patterns. The third objective is to integrate customer purchase prediction and market basket analysis to provide different product recommendations for different customer groups. This project tries to provide an integrated solution that can improve the customer satisfaction and produce revenue growth by using the strength of data analytics. The BG/NBD model is used to determine the optimal period that have highest prediction accuracy and it was determined that the six-month period is the optimal period with highest prediction accuracy. Then, the RFM segmentation categorizes customers into distinct groups such as where 50.74% of customers have been classified as "At risk," 46.98% as "Loyal customers", 1.62% as "Hibernating” and only 0.66% as “Champions”. Besides, both Apriori or the FP-growth algorithms are compared and find out that Apriori is faster for small datasets while FP-Growth is more efficient with large datasets due to its lower memory consumption. Thus, the FP-Growth algorithm is applied for each customer groups to perform Market Basket Analysis to discover frequent itemsets and provide product recommendations to each customer groups. The top-recommended items are different for each customer group with "PARTY BUNTING" being the most popular for both "Champions" and "Hibernating" customers, "REGENCY CAKESTAND 3 TIER" for "Loyal customers", and "WHITE HANGING HEART T-LIGHT HOLDER" for "At risk" customers. Furthermore, it was discovered that "At risk" customers generated fewer frequent itemsets, indicating less diverse purchase behaviour. In conclusion, this project is able to provide a better understanding of complex customer purchase patterns. The integration of those model offers practical tools to improve the customer engagement and enhance sales performance across various customer groups.

Actions (login required)