Predicting depression and unravelling its heterogeneous influences in middle-aged and older people populations: a machine learning approach

Table 2 Model performance in predicting courses of depressive symptoms for sensitivity analysis on only participants with complete data

Model	Accuracy	Sensitivity	PPV	Brier	AUC (95% CI)
Logistic regression	0.741	0.513	0.700	0.179	0.780 (0.775–0.784)
XGBOOST	0.740	0.514	0.699	0.179	0.781 (0.779–0.783)
Random forest	0.732	0.504	0.683	0.181	0.775 (0.770–0.778)
SVM	0.742	0.500	0.709	0.178	0.782 (0.776–0.786)
CNN	0.764	0.581	0.722	0.161	0.804 (0.782–0.809)

XGBOOST Extreme gradient boosting, SVM Support vector machine, CNN Convolutional neural network, PPV Positive predictive value, CI Confidence interval, parameter optimization: RF (max_depth = 10), SVM (kernel = linear), XGBOOST (n_estimators = 300, max_depth = 4, learning_rate = 0.01, gamma = 0.3, subsample = 0.7, colsample_bytree = 0.7), CNN (learning_rate = 0.003, epochs = 200, batch_size = 128)

ISSN: 2050-7283