Ant Colony-Partial Least Squares Algorithm (AOC_PLS): An Intelligent Variable Selection Method

The Ant Colony-Partial Least Squares Algorithm (AOC_PLS) is an intelligent variable selection method that integrates Ant Colony Optimization (ACO) with Partial Least Squares Regression (PLS). This algorithm addresses high-dimensional data challenges by simulating ant colony foraging behavior to search for optimal variable subsets, while utilizing PLS for dimensionality reduction to enhance modeling accuracy and computational efficiency. In AOC_PLS implementation, the ant colony algorithm explores the combinatorial space of variables. Each artificial ant represents a potential variable subset, selecting paths (variables) based on pheromone concentrations and heuristic information. High-quality subsets deposit stronger pheromone trails, thereby guiding subsequent search iterations. The Partial Least Squares component evaluates the predictive performance of variable subsets by extracting latent variables that maximize covariance between independent and dependent variables, effectively avoiding instability issues associated with traditional least squares methods under multicollinearity conditions. The MATLAB implementation of this algorithm is particularly suitable for scenarios where the number of variables significantly exceeds the number of samples (e.g., spectral data, genomic data). It effectively screens key variables while reducing overfitting risks through systematic feature selection. Key advantages include: 1) Global search capability prevents convergence to local optima; 2) PLS-based latent variable extraction preserves critical data characteristics; 3) Optimization of modeling for small-sample, high-dimensional datasets through intelligent variable weighting and selection mechanisms.