Abstract: Parkinson’s disease is a neurodegenerative disorder that inficts irreversible damage on humans. Some experimental data regarding Parkinson’s patients are redundant and irrelevant, posing signifcant challenges for disease detection. Therefore, there is a need to devise an efective method for the selective extraction of disease-specifc information, ensuring both accuracy and the utilization of fewer features. In this paper, a Binary Hybrid Artifcial Hummingbird and Flower Pollination Algorithm (FPA), called BFAHA, is proposed to solve the problem of Parkinson’s disease diagnosis based on speech signals. First, combining FPA with Artifcial Hummingbird Algorithm (AHA) can take advantage of the strong global exploration ability possessed by FPA to improve the disadvantages of AHA, such as premature convergence and easy falling into local optimum. Second, the Hemming distance is used to determine the diference between the other individuals in the population and the optimal individual after each iteration, if the diference is too signifcant, the cross-mutation strategy in the genetic algorithm (GA) is used to induce the population individuals to keep approaching the optimal individual in the random search process to speed up fnding the optimal solution. Finally, an S-shaped function converts the improved algorithm into a binary version to suit the characteristics of the feature selection (FS) tasks. In this paper, 10 high-dimensional datasets from UCI and the ASU are used to test the performance of BFAHA and apply it to Parkinson’s disease diagnosis. Compared with other state-of-the-art algorithms, BFAHA shows excellent competitiveness in both the test datasets and the classifcation problem, indicating that the algorithm proposed in this study has apparent advantages in the feld of feature selection.

Key words: Artifcial Hummingbird Algorithm , · Flower pollination algorithm , · Feature selection , · Parkinson’s disease , · Meta-heuristic