Predictive Modeling in Employee Satisfaction: An Approach with RA-SSA-SVM-Based Machine Learning Optimization

This paper presents RA-SSA-SVM, an innovative optimization framework for Support Vector Machines (SVM), addressing challenges in determining optimal parameter values, including the penalty parameter and the kernel function. Leveraging the Sea Squirt Algorithm (SSA) for global and local search, RA-SSA-SVM mitigates issues associated with conventional parameter tuning methods. The proposed RA-SSA algorithm, incorporating reflected backward learning and an adaptive control factor, prevents algorithmic stagnation in latestage SSA searches. Applied to an enterprise's regression prediction of employee satisfaction, the RA-SSA-SVM demonstrates enhanced convergence accuracy compared to traditional SSA. This research contributes to machine learning field by providing a solution to SVM parameter selection, emphasizing improved predictive modeling performance.

I. Introduction

In recent years, the Support Vector Machine (SVM) has gained recognition as a potent tool in machine learning for classification and regression tasks [1]. However, challenges arise in determining optimal parameter values, such as the penalty parameter and the choice of the kernel function, leading to suboptimal model performance and potential drawbacks.