A novel hybrid system for feature selection based on an improved gravitational search algorithm and k-NN method

• We propose an improved gravitational search algorithm which makes the best of ergodicity of piecewise linear chaotic map to explore the global search while utilizing the sequential quadratic programming to accelerate the local search.
• Based on the binary improved gravitational search algorithm and the k-nearest neighbor (k-NN) method, a novel hybrid system is proposed to improve classification accuracy with an appropriate feature subset in binary problems.
• The proposed system is able to select the discriminating input features correctly and achieve high classification accuracy which is comparable to or better than well-known similar classifier systems.

Feature selection is an important pre-processing step for solving classification problems. This problem is often solved by applying evolutionary algorithms in order to decrease the dimensional number of features involved. In this paper, we propose a novel hybrid system to improve classification accuracy with an appropriate feature subset in binary problems based on an improved gravitational search algorithm. This algorithm makes the best of ergodicity of piecewise linear chaotic map to explore the global search and utilizes the sequential quadratic programming to accelerate the local search. We evaluate the proposed hybrid system on several UCI machine learning benchmark examples, comparing our approaches with feature selection techniques and obtained better predictions with consistently fewer relevant features. Furthermore, the improved gravitational search algorithm is tested on 23 nonlinear benchmark functions and compared with 5 other heuristic algorithms. The obtained results confirm the high performance of the improved gravitational search algorithm in solving function optimization problems.

Figure optionsDownload as PowerPoint slide