Enhancing Data Collection for Market Basket Analysis Through CNN Object Detection

Analyzing data from night markets, particularly for market basket analysis, poses significant challenges. Previous studies collected images of products from customers at these markets, yet these images often depicted food items and products in plastic bags or boxes, complicating identification. To address this issue, our study utilizes object detection techniques to automatically label food items and products, thereby enhancing the accuracy and efficiency of market basket analysis. We evaluate the effectiveness of these techniques using standard evaluation metrics such as Precision-Recall curves and Average Precision (AP). Our results demonstrate commendable performance, with a mean Average Precision (mAP) score of 99.4% for food items without plastic bags and 99.3% for those with plastic bags. Notably, the combined model, capable of detecting both types of food items, achieves an mAP of 84.4%. Additionally, we utilized three association rule learning algorithms-Apriori, FP-Growth, and Eclat for market basket analysis to uncover meaningful associations among food categories. Among these algorithms, the Apriori algorithm produced the highest support value of 20% and confidence of 50%, generating a total of 8 rules. The accuracy of these association rules on a new dataset, comprising 20 transactions, is calculated to be 84%. These findings offer actionable insights for businesses in the food industry, empowering them to tailor marketing strategies and product offerings to better align with consumer needs and preferences. Ultimately, our study contributes to a deeper understanding of consumer behavior and product associations in the food industry, paving the way for future research endeavors in this domain.