A Hybrid Approach for Feature Selection and Weighting using Gravitational Search Algorithm in Churn Prediction

Customer retention is a critical concern for telecommunications companies, making accurate churn prediction models crucial for forecasting customer attrition. These models depend on large datasets with features of varying significance, underscoring the importance of effective feature selection and weighting to enhance prediction accuracy. This study proposes a novel hybrid approach that sequentially combines feature selection and weighting using the Gravitational Search Algorithm (GSA). By capitalizing on GSA's capacity to balance exploration and exploitation, the method first identifies the most relevant features and then assigns optimized weights to maximize their predictive contribution. The proposed FSFW-GSA method demonstrates superior performance compared to baseline models and existing GSA-based approaches, achieving notable improvements in accuracy, precision, recall, F1 score, and AUC. For example, FSFW-GSA attains an accuracy of 89.75%, an F1 score of 51.98%, and an AUC of 72.10%. By applying GSA in a structured two-step process, this study provides empirical evidence of its effectiveness in both reducing dimensionality and enhancing predictive performance.