Expression of genes found in the brains of autism, bipolar, and schizophrenia patients identified as overlapping. The overlap is a state in which the values of genes are similar. This paper aims to determine the best performance of support vector machines algorithm in classifying autism, bipolar, and schizophrenia based on the expression of genes using genome-wide association studies data. Using three support vector machine kernels, this study evaluates the performance of gaussian, laplacian, and sigmoid for genome-wide association studies datasets. The datasets were obtained from Psychiatric Genomics Consortium publications, where 660 data were taken with each disorder consisting of 220 data. This study proposes an optimal kernel for one-against-one and one-against-all multiclass support vector machine, and the performance is evaluated using accuracy. The study results show that the Gaussian kernel has the best accuracy performance compared to other support vector machines kernels in classifying genome-wide association studies data of autism, bipolar, and schizophrenia as early diagnosis.