Two-Stage Hybrid Feature Selection: Integrating ACO Algorithms With a Statistical Ensemble Technique for EV Demand Prediction

This paper presents a novel hybrid algorithm combining Ant Colony Optimization (ACO) and multi-model ensemble feature selection techniques to enhance the accuracy of Support Vector Regression (SVR) models for predicting Electric Vehicle Charging Demand (EVCD). Addressing the challenge of highly nonlinear data in EVCD time series, our approach integrates the exploratory strengths of ACO with the robustness of four statistical feature selection algorithms in a two-phased process. Initially, ACO identifies a preliminary subset of relevant features, which is then refined through a multi-model ensemble process to select the most relevant features via a vote of majority strategy. This process involves a thorough examination of a broad solution space to uncover promising feature subsets, acting as an effective method for dimensionality reduction and leading to a streamlined set of features for further analysis. Applied to a real-case study, the effectiveness of our framework was compared with that of basic multi-model, hybrid GA, and PSO multi-model techniques. The results underscore the superior efficiency of the proposed method.