Abstract:
"The retail industry faces the complex challenge of predicting individual customers' future
purchases, a task made difficult by the dynamic nature of consumer behaviour and the vast array of influencing factors. The CNSP project seeks to address this challenge by developing a predictive model that not only forecasts the items a customer is likely to buy in their next shopping trip but also anticipates the quantity of each item, aiming to enhance the personalisation of the shopping experience, optimise inventory management, and increase retail revenue.
The CNSP project employed a comprehensive approach combining data analytics and machine learning techniques to tackle this problem. The methodology involved collecting and analysing historical purchase data and customer demographics. A combination of two models, K Nearest Neighbour and Meta's Prophet, were developed to predict future shopping carts. The model's architecture was designed to be scalable and adaptable.
The initial deployment of the CNSP system demonstrated promising results, with the model achieving a 0.30 Jaccard coefficient score and an 0.33 f1 score in item prediction. The model exhibited robust performance for quantity prediction, as reflected in a Root Mean Square Error (RMSE) of 0.49, Mean Absolute Error (MAE) of 0.18 and a hit percentage of 0.85. These results suggest that the CNSP system can significantly enhance retail operations, though further validation and real-world testing are necessary to ascertain its efficacy and practical applicability fully."
