Hybrid grasshopper optimization algorithm with simulated annealing for feature selection using high dimensional dataset

Abstract

In the era of machine learning, microarray data play a crucial role in identifying cancer diseases. The impact of redundant and noisy features degrades the learning model’s performance. It may also increase the computational cost. The curse of dimensionality is the major concern in the case of microarray datasets. To eliminate this issue, feature selection methods play an effective role. This study proposes a hybrid filter-wrapper feature selection model using mRMR_Plus as a filter and grasshopper optimization algorithm as a wrapper. In the first stage of the proposed model, a ranked base filter mRMR_Plus is used to identify the top-ranked features from the original dataset. Cross-operator embedded simulated annealing (SA) is adopted to basic grasshopper optimization to develop a new wrapper model. The proposed model was tested with different cancer datasets to recognize the best optimal features. The result of mRMR-Plus-GO-SA is compared with different existing approaches. From the result and the comparative study, it’s noteworthy to state that the new mRMR_Plus-GO-SA filter wrapper model performs far better as compared to its counterparts in terms of the number of features selected and accuracy.