Mental Disorder Classification Prediction Based on AdaBoost Algorithm Optimized by Long Short-Term Memory Network
Year: 2024
Type: article
Abstract: This paper discusses the parameter optimization of AdaBoost algorithm through long short-term memory network (LSTM) to improve the performance of mental disorder classification model. This innovative combination not only enhances the model's ability to recognize complex patterns, but also improves its robustness to noise and outliers. We introduced the model in the experiment and observed that the accuracy curve showed that the model's prediction accuracy improved significantly from the initial 20% to nearly 100%, and remained stable at this level. At the same time, the value of the loss function also decreases from the initial 1.4 to the final 0.1, indicating that the model tends to converge and learn effectively. On the training set, through confusion matrix analysis, we found that the prediction results of manic bipolar disorder, depressive bipolar disorder, major depression and normal individuals were 100% accurate, indicating that the model performed well in the training stage. However, on the test set, while 27 patients were correctly classified, there were still eight patients who were incorrectly classified, resulting in a test set prediction accuracy of 77.14%. This shows that although the model performs very well on the training set, it still needs to be further optimized in practical applications to improve its adaptability to new data. In summary, this study demonstrates the positive effects of combining LSTM and AdaBoost algorithm, which provides an efficient and reliable method for classifying patients with mental disorders. Future work will focus on improving the model to enhance its ability to generalize in real-world applications to better serve clinical diagnosis and treatment.
Authors: Yucong Li, Yulin Jin
Institution: University of California, Berkeley
Language: English
Subfield: Health Information Management
Field: Health Professions
Domain: Health Sciences
Open Access status: hybrid
OpenAlex ID: W4405592193