Predicting Antenatal Depression during 3rd Trimester: A Machine Learning Approach with Feature Selection and TOPSIS Ranking

17 This study aimed to predict antenatal depression in erythroid women during their 3rd trimester. We investigated 18 prediction using four feature selection methods: Extra-tree classifier, Fisher score, and PCA. We also merged the 19 common features of the Extra-tree classifier and Fisher score and applied them to predicting antenatal depression in 20 the 3rd trimester of pregnancy. We gathered data from 62 women and their corresponding 18 attributes and evaluated 21 them using the Hamilton depression rating scale (HAM-D). Seven ML models were implemented to predict antenatal 22 depression, including k-nearest neighbors, Support vector machine, random forest, decision tree, bagging classifier, 23 multi-layer perception, and naïve Bayes. Therefore, the trained models were evaluated using various metrics, including 24 accuracy, sensitivity, specificity, precision, F1 score, FNR, FPR, and area under the receiver operating characteristic 25 curve. Ultimately all models were prioritized using TOPSIS with different feature selection methods, and the best 26 model was found to be DT without implementing any feature selection. The results of this study show the most 27 important factors in predicting depression in the 3rd trimester of pregnancy.