PREDICTIVE REGRESSION MODEL FOR SUBSCRIPTION FORECASTING THROUGH TELEMARKETING

Abstract

n this study, a data mining technique was employed to predict the success of telemarketing calls aimed at promoting long-term bank deposits, a key challenge for financial institutions aiming to optimize their marketing strategies. The dataset used for this analysis was sourced from a Portuguese retail bank, containing 45,211 records and 17 different attributes, including demographic and behavioral information about the clients. The primary objective was to develop a predictive model that could accurately determine whether a client would subscribe to a term deposit based on the features available in the dataset. To build the predictive model, logistic regression was applied, a statistical method well-suited for binary classification tasks like this one. The model was trained to estimate the probability that a client would respond positively to the marketing campaign. A crucial step in improving model performance was the feature selection process, where 22 distinct sets of features were evaluated. This helped identify the most relevant attributes that contributed to the prediction, ensuring that the final model was both efficient and interpretable. The final model achieved a precision of 0.74 and a recall of 0.74, indicating that it performed well in both identifying positive responses (precision) and capturing as many positive responses as possible (recall). Specifically, the model made 11,294 correct predictions, including 6,124 true positives and 5,170 true negatives, demonstrating its ability to accurately classify both successful and unsuccessful subscription cases. However, it also made 4,047 incorrect predictions, which consisted of 2,505 false positives and 1,542 false negatives. These errors reflect the inherent challenges in predictive modeling, particularly in distinguishing between clients who are likely to subscribe and those who are not.

Keywords

(Logistic Regression, Predictor, Bayes, Telemarketing Success, Customer Data Analysis, Socio-economic Factors, Deep Learning).