Project Tittle: BUILDING PERSONALIZED RECOMMENDATIONS WITH MACHINE LEARNING FOR FINANCIAL SERVICES MARKETING: A COLLABORATIVE FILTERING APPROACH
Summary Across the web-network as well as in physical stores, for customers, there is a huge number of choices that users are called to make decisions of preferences. This problem, along with the need of a company to promote a product or a service at the right time which eventually increases the company's revenues, indicates the necessity to filter, prioritise, and effectively provide related information to customers. Recommendation systems resolve these concerns by seeking thru big quantities of vigorously produced data to supply customers with customised content. A neighbourhood-based Collaborative Filtering model constitutes one of the filtering algorithms, well-known regarding their productiveness and simplicity.
This study investigates bank transaction data utilising Collaborative Filtering technique in order to make recommendations in accordance with customer tastes. The recommendations for an active customer are generated based on the shopping behaviour of similar customers. Additionally, all various merchants that appear in the dataset have been aggregated in certain, more general master categories. A comparison has been made between the volume and the value of the transaction across the merchant categories, a differentiation that led to the production of a new variable proportion of spend. Considering all together, the proportion of spend attribute will feed the algorithm in order to produce robust recommendations. Root Mean Squared Error (RMSE) has been selected as the measure of evaluation of the model, a measure that reveals how accurate are the recommendations made. Lastly, the problem exploration and analysis has been developed utilising the Cross-Industry Standard Process for Data Mining methodology.