<p>Feature selection is a challenging combinatorial optimization problem that seeks to identify informative feature subsets from high-dimensional data while balancing classification accuracy and model interpretability. Traditional metaheuristic algorithms often experience premature convergence due to insufficient control over population diversity and exploration dynamics, especially in the case of high-dimensional combinatorial optimization problems. To overcome this limitation, this paper proposes a chaotic Artificial Electric Field Algorithm (cAEFA), in which chaotic strategies are systematically embedded into Coulomb’s constant, a core parameter governing agent interactions in AEFA. Unlike random perturbations, chaotic maps introduce deterministic yet ergodic dynamics that generate controlled diversity and enable adaptive regulation of exploration and exploitation across different search stages. Different chaotic strategies, which differ in their nonlinear behaviours and sensitivity to initial conditions, are used, which allows the algorithm to exhibit varied search trajectories and enhanced robustness. A normalization mechanism is further included to stabilize these chaotic influences to ensure smooth convergence. Extensive experimental results on forty-two benchmark problems from the IEEE CEC test suites and real-world feature selection problems demonstrate that cAEFA achieves superior solution quality, faster convergence, and improved generalization compared with state-of-the-art methods. These findings highlight the effectiveness of chaos-driven parameter modulation as a principled mechanism for enhancing metaheuristic search performance. The MATLAB code of cAEFA can be found at <a href="https://github.com/ChauhanDikshit">https://github.com/ChauhanDikshit</a>.</p>