Integration of Kestrel-based search algorithm with artificial neural network for feature subset selection
Israel Edem Agbehadji, Richard Charles Millham, Simon James Fong, Hongji Yang · International Journal of Bio-Inspired Computation · 2019
Feature selection plays an important role in data pre-processing of data management. Although there are different methods available for feature selection such as filter, wrapper and embedded methods, selecting relevant features still remains a challenge in the current dispensation of big data. This paper proposes a new meta-heuristic method that integrates with wrapper method for feature subset selection. A mathematical model is formulated using random encircling and imitative behaviour (REIM) of the Kestrel bird for optimal selection of features. A test dataset from a benchmark was used to test the proposed algorithm. The performance of proposed algorithm was evaluated against PSO and ACO. The proposed model is observed to provide low error rate of 0.001143 as compared with PSO (0.0589) and ACO (0.05236). In terms of optimal size over dimension of each dataset, the proposed model performed well in 3 out of 4 datasets, while PSO-ANN performed well in 1 out of 4 datasets, ACO-ANN could not perform in any of the dataset.