Optimal column subset selection for image classification by genetic algorithms

摘要: Many problems in operations research can be solved by combinatorial optimization. Fixed-length subset selection is a family of optimization that involve set unique objects from larger superset. Feature selection, p-median problem, and column problem are three examples hard search for fixed-length subsets. Due to their high complexity, exact algorithms often infeasible solve real-world instances these approximate methods based on various heuristic metaheuristic (e.g. nature-inspired) approaches employed. Selecting subsets massive data matrices an important technique useful construction compressed representations low rank approximations high-dimensional data. Search optimal exactly k columns matrix, \(A^{m\times n}\), \(k < n\), well-known with practical implications processing mining. It used unsupervised feature dimensionality reduction, visualization, so on. A representation raw contribute, example, reduction algorithm training times supervised learning, elimination overfitting classification regression, facilitation better understanding, many other benefits. This paper proposes novel genetic the evaluates it series computational experiments image classification. The evaluation shows proposed modifications improve results obtained artificial evolution.