Potential Customers Prediction in Bank Telemarketing

Abstract

Data mining plays a vital role in the success of direct marketing campaigns by predicting which leads subscribe to a term deposit. This thesis is accomplished to illustrate with practical mining methods that the data is related to a Portuguese banking institution's direct marketing campaigns (phone calls). The algorithms are used: K-Nearest Neighbor, Logistic Regression, Linear Supported Vector Machines, and Extreme Gradient Boosting to classify potential customers for long-term deposits finance products. Response coding is used to vectorize categorical data while solving a machine learning classification problem. Accuracy and AUC scores are key metrics to evaluate performance. We inherited selecting important features from previous research. Our thesis employed a better method by combining response coding techniques with practical algorithms in an unbalanced dataset. The best prediction model achieved 91.07% and 0.9324 of accuracy and AUC score, significantly higher than the prior of 79% and 0.8 respectively