Judgment of Alzheemer’s Desease Based on Multi-feature Mixed Model

In order to solve the problems of complex, time-consuming and expensive diagnostic methods and processes of Alzheimer's disease (AD), a multi feature fusion ad judgment model is proposed in this paper. According to the voice information of the tested person, MFCC feature extraction and text data conversion are carried out respectively. The overall integration of MFCC feature, artificial audio feature and artificial text feature is realized by integrating audio feature and text feature, and the number of model layers is compressed to 50% of the original through the Bert model of knowledge distillation. The experimental results show that the accuracy of the model after multi feature fusion is 83.69%, while the highest prediction accuracy of the single feature model is only 77.78%. The running time of the compressed model on the computer is about 20 minutes, which is about 30 minutes less than the original model.