A novel methodological approach to SaaS churn prediction using whale optimization algorithm

Abstract

Customer churn is a critical concern in the Software as a Service (SaaS) sector, potentially impacting long-term growth within the cloud computing industry. The scarcity of research on customer churn models in SaaS, particularly regarding diverse feature selection methods and predictive algorithms, highlights a significant gap. Addressing this would enhance academic discourse and provide essential insights for managerial decision-making. This study introduces a novel approach to SaaS churn prediction using the Whale Optimization Algorithm (WOA) for feature selection. Results show that WOA-reduced datasets improve processing efficiency and outperform full-variable datasets in predictive performance. The study encompasses a range of prediction techniques with three distinct datasets evaluated derived from over 1,000 users of a multinational SaaS company: the WOA-reduced dataset, the full-variable dataset, and the chi-squared-derived dataset. These three datasets were examined with the most used in literature, k-nearest neighbor, Decision Trees, Naïve Bayes, Random Forests, and Neural Network techniques, and the performance metrics such as Area Under Curve, Accuracy, Precision, Recall, and F1 Score were used as classification success. The results demonstrate that the WOA-reduced dataset outperformed the full-variable and chi-squared-derived datasets regarding performance metrics.