Multi-objective Harris Hawk Algorithms for the Diagnosis of Parkinson's Disease

Abstract

This study proposes new binary Harris Hawk Optimization algorithms for diagnosing Parkinson's disease. New exploration and exploitation operators are developed, and a K-Nearest Neighbour classifier that adapts to the given dataset is employed. A parallel version of the algorithm implemented using Message Passing Interfaces is proposed for large problem instances where the fitness evaluation is time-consuming. Comparisons with state-of-the-art genetic, particle swarm, binary bat, cuckoo search, and grey wolf algorithms verified that our proposed algorithms are the best methods in the literature on average in terms of prediction accuracy values. In total, a 32.5% reduction is achieved in the number of features of all datasets. We report new best solutions for the first time in literature. For three datasets out of four, we outperformed state-of-the-art algorithms.