Research on the detection method of mental dysfunction of EEG signals based on multi-model fusion

Early diagnosis of psychiatric disorders faces the challenges of high misdiagnosis rate of traditional methods and the susceptibility of electroencephalogram (EEG) signals to interference, to this end, this study proposes a multi-model fusion-based EEG signal analysis method, which constructs a stacked fusion model by improving convolutional neural network (CNN) and combining with support vector machines (SVMs) and random forests (RFs) to take advantage of the complementary nature of the frequency-domain and spatio-temporal features, and at the same time PCA dimensionality reduction and cross-validation are used to optimise the feature expression and generalisation capability. Experimental results show that the method achieves 86.5% classification accuracy (AUC=0.927) on the simulated EEG dataset, which is superior to that of a single model, innovatively breaks through the performance bottleneck of a single model, and supports real-time analysis and clinical deployment through a lightweight design, providing a more reliable solution for the diagnosis of mental disorders.