Enchancing Sequential Floating Search Feature Selection using Niching-Genetic Algorithm

Classification, a process for predicting the class of a given input data, is one of the most fundamental tasks in data mining. Feature Selection is a crucial and beneficial task for data preparation of data modelling because its process to choose a small subset of relevant features from the original ones by eliminating noisy, irrelevant and redundant. We applied ten popular datasets combining smaller and higher number of variables such as Breast Cancer, Iris, Abalone and Heart datasets to conduct the experiment. Combining niching algorithms with feature selection is inspired by the need to avoid local optima and maintain diversity in feature subsets. Niching helps explore different areas of the solution space, increasing the chances of finding the global optimum. We provide results for deterministic crowding; show how different crowding replacement rules can be combined in algorithms. The experimental results show that SVM classifier gives the highest accuracy and low error rate. In probabilistic crowding, subpopulations are maintained stably, and we demonstrate that it is possible to analyze and bring the new sub features that genetic algorithm can't discover. We also provide novel results for deterministic crowding, show how different crowding replacement rules can be combined in portfolios, and discuss population sizing.