Multi-classifier fusion base on belief-value for the diagnosis of neuropsychiatric disorders

Abstract Neuropsychiatric disorders seriously affect the health of patients, and early diagnosis and treatment are crucial to improve the quality of patients’ life. Machine learning and other related methods can be used for disease diagnosis and prediction, among which multi-classifier fusion method has been widely studied due to its significant performance over single classifiers. In this paper, we propose a multi-classifier fusion classification framework based on belief-valuefor the neuropsychiatric disorders diagnosis. Specifically, the belief-value measures the belief level of different samples by considering information from two perspectives, which are distance information (the output distance of the classifier) and local density information (the weight of the nearest neighbor samples on the test samples). The proposed belief-value is more representative compared to the belief-value which only uses a single type of information. Further, based on the concept of multi-view learning, we performed the calculation of the belief-values under the sample space with different features, and the complementary relationship between different belief-values was captured by a multilayer perceptual (MLP) network. Compared with majority voting and linear fusion methods, the MLP network can better capture the nonlinear relationship between belief-values, which produces better diagnostic results. Experimental results show that the proposed method outperforms single classifier and multi-classifier linear fusion methods for the diagnosis of neuropsychiatric disorders.