Deep learning based depression classification using environmental factor selection

Previous studies have examined whether symptoms found in annual health examinations could be predictive to classify a person with depressive disorders. In this paper, Convolutional Neural Network(CNN) and Light Gradient Boosting Machine (LightGBM)-based depression classification models were proposed based on physical and environmental information of health examinations. For this, input data of CNN and LightGBM were pre-processed by adding and excluding several environmental information that could highly affect the prediction results. And the optimal model of CNN and LightGBM were obtained through hyperparameter analysis to maximize the depression classification performance. Performance results proved that the predictive accuracy of 2D-CNN was 78.71% and AUC values for 1D-CNN, 2D-CNN, LightGBM were 0.750, 0.716, 0.731, respectively. By comparing performance results, our proposed classification models outperformed the ANN and DNN-based conventional models in terms of accuracy and AUC. Copyright © The Korean Institute of Electrical Engineers