/Users/yuchi/PycharmProjects/PsyMl_ISI/.venv/bin/python /Users/yuchi/PycharmProjects/PsyMl_ISI/ML/ml_benchmark_modular.py
[資料] 來源=isi_raw_data 目標=3TP 列數=51 特徵數=4
[特徵] 使用欄位（前 15）：['BDI_T1', 'BAI_T1', 'CD_RISC_T1', 'DBAS']
[CV] Stratified 5-fold, seed=42 | class_weight=balanced
[聚合] K-fold = weighted | LOSO = weighted
[Leakage check] Class balance
count percent%
3TP
0 35 68.6
1 16 31.4
[Leakage check] 未發現與目標 |r| ≥ 0.95 的欄位。
=== Basic ML Benchmark (Stratified 5-fold CV) ===
model AUC F1_pos(=1) Prec_pos Rec_pos F1_neg(=0) Prec_neg Rec_neg MCC Accuracy Pred1_mean Pred0_mean
NaiveBayes 0.921 0.759 0.846 0.688 0.904 0.868 0.943 0.671 0.863 2.600 7.600
LogisticRegression 0.909 0.800 0.857 0.750 0.917 0.892 0.943 0.720 0.882 2.800 7.400
KNN 0.904 0.759 0.846 0.688 0.904 0.868 0.943 0.671 0.863 2.600 7.600
SVM 0.891 0.848 0.824 0.875 0.928 0.941 0.914 0.777 0.902 3.400 6.800
RandomForest 0.860 0.710 0.733 0.688 0.873 0.861 0.886 0.584 0.824 3.000 7.200
MLP 0.852 0.634 0.520 0.812 0.754 0.885 0.657 0.436 0.706 5.000 5.200
XGBoost 0.791 0.710 0.733 0.688 0.873 0.861 0.886 0.584 0.824 3.000 7.200
DecisionTree 0.664 0.533 0.571 0.500 0.806 0.784 0.829 0.342 0.725 2.800 7.400
--- Aggregated Confusion Matrix Sums (across all folds' test parts) ---
model TN_sum FP_sum FN_sum TP_sum FP_FN_IDS
NaiveBayes 33 2 5 11 [S112194, S112019 | S112002, S112008, S112036, S112169, S112183]
LogisticRegression 33 2 4 12 [S112194, S112019 | S112002, S112008, S112169, S112183]
KNN 33 2 5 11 [S112194, S112019 | S112002, S112008, S112036, S112169, S112183]
SVM 32 3 2 14 [S112194, S112019, S112104 | S112008, S112169]
RandomForest 31 4 5 11 [S112194, S112019, S112012, S112105 | S112008, S112115, S112036, S112169, S112183]
MLP 23 12 3 13 [S112194, S112016, S112019, S112159, S112171, S112105, S112164, S112234, S112043, S112070, S112104, S112186 | S112008, S112086, S112115]
XGBoost 31 4 5 11 [S112194, S112019, S112012, S112105 | S112008, S112115, S112036, S112169, S112183]
DecisionTree 29 6 8 8 [S112194, S112019, S112012, S112105, S112186, S112176 | S112002, S112008, S112115, S112039, S112036, S112169, S112183, S112029]
[SKIP] LOSO 已停用（RUN_LOSO=False）
/Users/yuchi/PycharmProjects/PsyMl_ISI/.venv/bin/python /Users/yuchi/PycharmProjects/PsyMl_ISI/ML/ml_benchmark_modular.py
[資料] 來源=isi_raw_data 目標=3TP 列數=51 特徵數=3
[特徵] 使用欄位（前 15）：['BDI_T1', 'CD_RISC_T1', 'DBAS']
[CV] Stratified 5-fold, seed=42 | class_weight=balanced
[聚合] K-fold = weighted | LOSO = weighted
[Leakage check] Class balance
count percent%
3TP
0 35 68.6
1 16 31.4
[Leakage check] 未發現與目標 |r| ≥ 0.95 的欄位。
=== Basic ML Benchmark (Stratified 5-fold CV) ===
model AUC F1_pos(=1) Prec_pos Rec_pos F1_neg(=0) Prec_neg Rec_neg MCC Accuracy Pred1_mean Pred0_mean
KNN 0.925 0.786 0.917 0.688 0.919 0.872 0.971 0.721 0.882 2.400 7.800
NaiveBayes 0.914 0.800 0.857 0.750 0.917 0.892 0.943 0.720 0.882 2.800 7.400
LogisticRegression 0.912 0.765 0.722 0.812 0.882 0.909 0.857 0.650 0.843 3.600 6.600
SVM 0.895 0.750 0.750 0.750 0.886 0.886 0.886 0.636 0.843 3.200 7.000
MLP 0.868 0.316 1.000 0.188 0.843 0.729 1.000 0.370 0.745 0.600 9.600
RandomForest 0.864 0.710 0.733 0.688 0.873 0.861 0.886 0.584 0.824 3.000 7.200
XGBoost 0.796 0.688 0.688 0.688 0.857 0.857 0.857 0.545 0.804 3.200 7.000
DecisionTree 0.693 0.571 0.667 0.500 0.838 0.795 0.886 0.422 0.765 2.400 7.800
--- Aggregated Confusion Matrix Sums (across all folds' test parts) ---
model TN_sum FP_sum FN_sum TP_sum FP_FN_IDS
KNN 34 1 5 11 [S112019 | S112002, S112008, S112169, S112183, S112029]
NaiveBayes 33 2 4 12 [S112194, S112019 | S112002, S112008, S112169, S112183]
LogisticRegression 30 5 3 13 [S112194, S112019, S112012, S112105, S112070 | S112002, S112008, S112183]
SVM 31 4 4 12 [S112194, S112019, S112105, S112104 | S112002, S112008, S112169, S112183]
MLP 35 0 13 3 [- | S112002, S112003, S112074, S112192, S112008, S112039, S112055, S112087, S112036, S112169, S112183, S112029, S112075]
RandomForest 31 4 5 11 [S112194, S112019, S112012, S112105 | S112003, S112008, S112115, S112169, S112029]
XGBoost 30 5 5 11 [S112194, S112019, S112012, S112105, S112119 | S112008, S112115, S112036, S112169, S112029]
DecisionTree 31 4 8 8 [S112194, S112019, S112012, S112105 | S112002, S112008, S112115, S112039, S112036, S112169, S112183, S112029]
[SKIP] LOSO 已停用（RUN_LOSO=False）
[資料] 來源=isi_raw_data 目標=3TP 列數=51 特徵數=4
[特徵] 使用欄位（前 15）：['BDI_T1', 'CD_RISC_T1', 'DBAS', 'HRV_VLF']
[CV] Stratified 5-fold, seed=42 | class_weight=balanced
[聚合] K-fold = weighted | LOSO = weighted
[Leakage check] Class balance
count percent%
3TP
0 35 68.6
1 16 31.4
[Leakage check] 未發現與目標 |r| ≥ 0.95 的欄位。
=== Basic ML Benchmark (Stratified 5-fold CV) ===
model AUC F1_pos(=1) Prec_pos Rec_pos F1_neg(=0) Prec_neg Rec_neg MCC Accuracy Pred1_mean Pred0_mean
NaiveBayes 0.920 0.800 0.857 0.750 0.917 0.892 0.943 0.720 0.882 2.800 7.400
KNN 0.901 0.692 0.900 0.562 0.895 0.829 0.971 0.624 0.843 2.000 8.200
SVM 0.896 0.788 0.765 0.812 0.899 0.912 0.886 0.687 0.863 3.400 6.800
LogisticRegression 0.891 0.812 0.812 0.812 0.914 0.914 0.914 0.727 0.882 3.200 7.000
RandomForest 0.880 0.621 0.692 0.562 0.849 0.816 0.886 0.477 0.784 2.600 7.600
XGBoost 0.812 0.667 0.714 0.625 0.861 0.838 0.886 0.531 0.804 2.800 7.400
DecisionTree 0.633 0.483 0.538 0.438 0.795 0.763 0.829 0.283 0.706 2.600 7.600
MLP 0.600 0.452 0.467 0.438 0.761 0.750 0.771 0.213 0.667 3.000 7.200
--- Aggregated Confusion Matrix Sums (across all folds' test parts) ---
model TN_sum FP_sum FN_sum TP_sum FP_FN_IDS
NaiveBayes 33 2 4 12 [S112194, S112019 | S112002, S112008, S112169, S112183]
KNN 34 1 7 9 [S112019 | S112002, S112003, S112008, S112086, S112169, S112183, S112029]
SVM 31 4 3 13 [S112016, S112019, S112158, S112104 | S112002, S112008, S112169]
LogisticRegression 32 3 3 13 [S112019, S112105, S112070 | S112002, S112008, S112169]
RandomForest 31 4 7 9 [S112194, S112019, S112012, S112105 | S112002, S112003, S112008, S112115, S112036, S112169, S112183]
XGBoost 31 4 6 10 [S112194, S112019, S112012, S112105 | S112002, S112003, S112008, S112115, S112036, S112169]
DecisionTree 29 6 9 7 [S112194, S112019, S112012, S112105, S112070, S112186 | S112002, S112003, S112008, S112115, S112039, S112036, S112169, S112183, S112029]
MLP 27 8 9 7 [S112079, S112082, S112173, S112239, S112047, S112171, S112042, S112119 | S112002, S112003, S112192, S112008, S112036, S112169, S112183, S112023, S112029]
[SKIP] LOSO 已停用（RUN_LOSO=False）
Tip
⭐ AUC提升分析
Baseline (BDI+BAI): 0.878
加入HRV : 0.857 (-0.021, -2.4%) ❌ 反而下降
加入CD-RISC+DBAS : 0.921 (+0.043, +4.9%) ✅ 顯著提升
移除BAI : 0.925 (+0.047, +5.4%) ✅ 最佳效果
CD-RISC+DBAS+HRV : 0.920 (+0.042, +4.8%) ✅ 略降但仍優於baseline
18 24 25