HRV狀態差異分析：
-
嘗試用HRV在各個狀態時的差異來跑ML
-
焦慮反應與恢復（A1-A3）
HRV_DeltaAnx_RMSSD= RMSSD(A2) − RMSSD(A1)HRV_RecAnx_RMSSD= RMSSD(A3) − RMSSD(A2)HRV_ReturnAnx_RMSSD= RMSSD(A3) − RMSSD(A1)
-
注意力負荷反應（A6-A7）
HRV_DeltaAttn_MeanNN= MeanNN(A7) − MeanNN(A6)HRV_DeltaAttn_RMSSD= RMSSD(A7) − RMSSD(A6)HRV_DeltaAttn_SD1= SD1(A7) − SD1(A6)
-
正向情緒恢復（A4-A5 + 回到基線）
HRV_RecPos_RMSSD= RMSSD(A5) − RMSSD(A4)HRV_ReturnPos_RMSSD= RMSSD(A5) − RMSSD(A1)HRV_ReturnPos_MeanNN= MeanNN(A5) − MeanNN(A1)
傳統統計結果(normal ranking)：
/Users/yuchi/PycharmProjects/PsyMl_Data/.venv/bin/python /Users/yuchi/PycharmProjects/PsyMl_ISI/ML/tools/normal_ranking.py
========================================================================
[3TP] 來源：isi_raw_data_hrv_recalc（已在 SQL where 過濾 3TP IS NOT NULL）
[概況] 列=33 欄=1026 總缺值=10338
[目標 '3TP'] 樣本數：33
- 類別 0: 19 (57.6%)
- 類別 1: 14 (42.4%)
------------------------------------------------------------------------
[3TP] 模式=允許 | 允許欄位數=9
- 實際可用特徵數：9
- 排序依據：info_gain | Top 9
info_gain gain_ratio SU gini_decrease anova_f chi2
HRV_DeltaAttn_MeanNN 0.528 0.265 0.354 0.037 0.278 14.538
HRV_ReturnPos_MeanNN 0.455 0.181 0.259 0.000 0.040 12.555
HRV_ReturnPos_RMSSD 0.410 0.188 0.259 0.000 0.857 11.215
HRV_ReturnAnx_RMSSD 0.352 0.172 0.232 0.000 0.008 9.788
HRV_RecAnx_RMSSD 0.277 0.115 0.163 0.000 0.000 8.100
HRV_DeltaAttn_RMSSD 0.257 0.146 0.187 0.247 0.282 7.003
HRV_DeltaAttn_SD1 0.257 0.146 0.187 0.128 0.264 7.003
HRV_RecPos_RMSSD 0.250 0.089 0.131 0.416 1.557 7.762
HRV_DeltaAnx_RMSSD 0.203 0.153 0.175 0.172 0.013 6.057
进程已结束，退出代码为 0
Lasso Ranking
/Users/yuchi/PycharmProjects/PsyMl_Data/.venv/bin/python /Users/yuchi/PycharmProjects/PsyMl_ISI/ML/tools/lasso_ranking.py
[模式] 允許模式
[資料] 來源=isi_raw_data_hrv_recalc 目標=3TP 樣本=33 特徵=9
已排除群組：['ACS', 'CPT', 'EEG', 'IGT', 'ISI', 'PSQI', 'WM']
總缺值比例：18.18%
[CV結果]（分數 = neg_log_loss，越大越好 → log_loss 越小）
lambda_min：C = 0.199635 | lambda = 5.00913
lambda_2SE：C = 0.0316228 | lambda = 31.6228
[選入變項（2SE）] 以 |係數| 排序（前 30）
（無變項被選入；可放寬正則或檢查特徵）
[對照] lambda_min 非零變項數：0，lambda_2SE 非零變項數：0
[Top 10（路徑峰值）] 不綁定單一 C
HRV_ReturnPos_MeanNN
HRV_DeltaAttn_RMSSD
HRV_DeltaAttn_MeanNN
HRV_RecPos_RMSSD
HRV_ReturnPos_RMSSD
HRV_DeltaAnx_RMSSD
HRV_RecAnx_RMSSD
HRV_ReturnAnx_RMSSD
HRV_DeltaAttn_SD1
Lasso Ranking 200次循環
/Users/yuchi/PycharmProjects/PsyMl_Data/.venv/bin/python /Users/yuchi/PycharmProjects/PsyMl_ISI/ML/tools/lasso_ranking/1_lasso_stability_ranking.py
[資料] 來源=isi_raw_data_hrv_recalc 目標=3TP 樣本=33 特徵=9
[穩定性排名] 已輸出：/Users/yuchi/PycharmProjects/PsyMl_ISI/ML/tools/outputs/lasso_stability_ranking.csv
feature count freq mean_abs_coef
HRV_DeltaAttn_SD1 200 1.0 10.310658
HRV_DeltaAttn_RMSSD 200 1.0 5.253410
HRV_RecPos_RMSSD 200 1.0 2.105531
HRV_DeltaAttn_MeanNN 200 1.0 1.630612
HRV_DeltaAnx_RMSSD 200 1.0 1.394721
HRV_ReturnPos_RMSSD 200 1.0 1.328560
HRV_ReturnPos_MeanNN 200 1.0 1.257333
HRV_ReturnAnx_RMSSD 200 1.0 0.692708
HRV_RecAnx_RMSSD 200 1.0 0.630162
運算結果
- 僅
BAI_T1,BDI_T1
[資料] 目標=3TP 樣本=27 特徵=2
使用欄位：['BDI_T1', 'BAI_T1']
[SMOTE] OFF (mode=Standard, k_neighbors=5, class_weight_when_smote=True)
[CLASS_WEIGHT] mode=Off
[目標分佈]
- 類別 0: 15 (55.6%)
- 類別 1: 12 (44.4%)
[偵測] 目標型態：binary；classes=[0 1]
=== Basic ML Benchmark (Standard | Repeated Stratified 5-fold x 100 CV) | seed=42 | SMOTE=OFF | class_weight=OFF ===
model AUC F1_pos(=1) Prec_pos Rec_pos F1_neg(=0) Prec_neg Rec_neg MCC Accuracy Pred1_mean Pred0_mean
NaiveBayes 0.942 0.737 0.855 0.696 0.847 0.824 0.915 0.653 0.820 1.936 3.464
LogisticRegression 0.938 0.719 0.848 0.664 0.846 0.804 0.925 0.630 0.812 1.830 3.570
RandomForest 0.912 0.678 0.732 0.690 0.771 0.798 0.795 0.514 0.749 2.272 3.128
SVM 0.895 0.705 0.823 0.663 0.829 0.800 0.898 0.601 0.796 1.906 3.494
KNN 0.880 0.681 0.765 0.673 0.806 0.809 0.855 0.564 0.775 2.054 3.346
DecisionTree 0.738 0.679 0.741 0.686 0.764 0.791 0.791 0.510 0.745 2.278 3.122
--- Aggregated Confusion Matrix Sums (Standard) ---
model TN_sum FP_sum FN_sum TP_sum
NaiveBayes 1372 128 360 840
LogisticRegression 1387 113 398 802
RandomForest 1193 307 371 829
SVM 1347 153 400 800
KNN 1283 217 390 810
DecisionTree 1186 314 375 825
--- Specificity / Sensitivity (Standard) ---
model Specificity Sensitivity
NaiveBayes 0.915 0.700
LogisticRegression 0.925 0.668
RandomForest 0.795 0.691
SVM 0.898 0.667
KNN 0.855 0.675
DecisionTree 0.791 0.688
- 加上
HRV_DeltaAttn_MeanNN
[資料] 目標=3TP 樣本=27 特徵=3
使用欄位：['BDI_T1', 'BAI_T1', 'HRV_DeltaAttn_MeanNN']
[SMOTE] OFF (mode=Standard, k_neighbors=5, class_weight_when_smote=True)
[CLASS_WEIGHT] mode=Off
[目標分佈]
- 類別 0: 15 (55.6%)
- 類別 1: 12 (44.4%)
[偵測] 目標型態：binary；classes=[0 1]
=== Basic ML Benchmark (Standard | Repeated Stratified 5-fold x 100 CV) | seed=42 | SMOTE=OFF | class_weight=OFF ===
model AUC F1_pos(=1) Prec_pos Rec_pos F1_neg(=0) Prec_neg Rec_neg MCC Accuracy Pred1_mean Pred0_mean
SVM 0.912 0.728 0.853 0.674 0.856 0.811 0.932 0.645 0.820 1.828 3.572
RandomForest 0.902 0.709 0.785 0.705 0.802 0.817 0.839 0.579 0.780 2.180 3.220
LogisticRegression 0.875 0.711 0.842 0.657 0.842 0.799 0.921 0.619 0.806 1.820 3.580
KNN 0.854 0.696 0.832 0.638 0.846 0.794 0.933 0.611 0.805 1.738 3.662
NaiveBayes 0.842 0.709 0.822 0.674 0.822 0.804 0.885 0.601 0.794 1.974 3.426
DecisionTree 0.727 0.658 0.719 0.667 0.756 0.778 0.788 0.486 0.734 2.236 3.164
--- Aggregated Confusion Matrix Sums (Standard) ---
model TN_sum FP_sum FN_sum TP_sum
SVM 1398 102 388 812
RandomForest 1258 242 352 848
LogisticRegression 1382 118 408 792
KNN 1400 100 431 769
NaiveBayes 1327 173 386 814
DecisionTree 1182 318 400 800
--- Specificity / Sensitivity (Standard) ---
model Specificity Sensitivity
SVM 0.932 0.677
RandomForest 0.839 0.707
LogisticRegression 0.921 0.660
KNN 0.933 0.641
NaiveBayes 0.885 0.678
DecisionTree 0.788 0.667
- 加上
HRV_ReturnPos_MeanNN
[資料] 目標=3TP 樣本=27 特徵=3
使用欄位：['BDI_T1', 'BAI_T1', 'HRV_ReturnPos_MeanNN']
[SMOTE] OFF (mode=Standard, k_neighbors=5, class_weight_when_smote=True)
[CLASS_WEIGHT] mode=Off
[目標分佈]
- 類別 0: 15 (55.6%)
- 類別 1: 12 (44.4%)
[偵測] 目標型態：binary；classes=[0 1]
=== Basic ML Benchmark (Standard | Repeated Stratified 5-fold x 100 CV) | seed=42 | SMOTE=OFF | class_weight=OFF ===
model AUC F1_pos(=1) Prec_pos Rec_pos F1_neg(=0) Prec_neg Rec_neg MCC Accuracy Pred1_mean Pred0_mean
NaiveBayes 0.911 0.765 0.840 0.752 0.840 0.847 0.873 0.658 0.821 2.194 3.206
LogisticRegression 0.898 0.711 0.838 0.657 0.844 0.801 0.922 0.618 0.807 1.818 3.582
RandomForest 0.890 0.639 0.730 0.631 0.772 0.775 0.824 0.492 0.739 2.042 3.358
SVM 0.877 0.752 0.844 0.727 0.842 0.832 0.889 0.650 0.818 2.084 3.316
KNN 0.868 0.670 0.813 0.611 0.832 0.779 0.927 0.580 0.788 1.686 3.714
DecisionTree 0.718 0.640 0.704 0.638 0.756 0.765 0.797 0.463 0.727 2.140 3.260
--- Aggregated Confusion Matrix Sums (Standard) ---
model TN_sum FP_sum FN_sum TP_sum
NaiveBayes 1309 191 294 906
LogisticRegression 1383 117 408 792
RandomForest 1236 264 443 757
SVM 1333 167 325 875
KNN 1390 110 467 733
DecisionTree 1196 304 434 766
--- Specificity / Sensitivity (Standard) ---
model Specificity Sensitivity
NaiveBayes 0.873 0.755
LogisticRegression 0.922 0.660
RandomForest 0.824 0.631
SVM 0.889 0.729
KNN 0.927 0.611
DecisionTree 0.797 0.638
DecisionTree 0.798 0.655
- 加上
HRV_RecPos_RMSSD
[資料] 目標=3TP 樣本=27 特徵=3
使用欄位：['BDI_T1', 'BAI_T1', 'HRV_RecPos_RMSSD']
[SMOTE] OFF (mode=Standard, k_neighbors=5, class_weight_when_smote=True)
[CLASS_WEIGHT] mode=Off
[目標分佈]
- 類別 0: 15 (55.6%)
- 類別 1: 12 (44.4%)
[偵測] 目標型態：binary；classes=[0 1]
=== Basic ML Benchmark (Standard | Repeated Stratified 5-fold x 100 CV) | seed=42 | SMOTE=OFF | class_weight=OFF ===
model AUC F1_pos(=1) Prec_pos Rec_pos F1_neg(=0) Prec_neg Rec_neg MCC Accuracy Pred1_mean Pred0_mean
LogisticRegression 0.937 0.745 0.855 0.702 0.861 0.828 0.928 0.665 0.830 1.906 3.494
RandomForest 0.901 0.687 0.775 0.672 0.802 0.799 0.850 0.557 0.772 2.066 3.334
SVM 0.893 0.768 0.876 0.724 0.866 0.837 0.927 0.688 0.839 1.960 3.440
KNN 0.882 0.658 0.813 0.590 0.827 0.767 0.928 0.563 0.780 1.638 3.762
NaiveBayes 0.842 0.725 0.852 0.674 0.842 0.808 0.917 0.633 0.812 1.880 3.520
DecisionTree 0.718 0.636 0.705 0.632 0.757 0.763 0.804 0.465 0.729 2.108 3.292
--- Aggregated Confusion Matrix Sums (Standard) ---
model TN_sum FP_sum FN_sum TP_sum
LogisticRegression 1392 108 355 845
RandomForest 1275 225 392 808
SVM 1391 109 329 871
KNN 1392 108 489 711
NaiveBayes 1375 125 385 815
DecisionTree 1206 294 440 760
--- Specificity / Sensitivity (Standard) ---
model Specificity Sensitivity
LogisticRegression 0.928 0.704
RandomForest 0.850 0.673
SVM 0.927 0.726
KNN 0.928 0.593
NaiveBayes 0.917 0.679
DecisionTree 0.804 0.633
結合HRV_DeltaAttn_MeanNN, HRV_RecPos_RMSSD
[目標分佈]
- 類別 0: 15 (55.6%)
- 類別 1: 12 (44.4%)
[偵測] 目標型態：binary；classes=[0 1]
=== Basic ML Benchmark (Standard | Repeated Stratified 5-fold x 100 CV) | seed=42 | SMOTE=OFF | class_weight=OFF ===
model AUC F1_pos(=1) Prec_pos Rec_pos F1_neg(=0) Prec_neg Rec_neg MCC Accuracy Pred1_mean Pred0_mean
SVM 0.916 0.807 0.890 0.778 0.885 0.867 0.929 0.736 0.863 2.078 3.322
RandomForest 0.899 0.691 0.773 0.682 0.797 0.803 0.841 0.558 0.771 2.120 3.280
KNN 0.866 0.651 0.809 0.580 0.829 0.762 0.933 0.557 0.779 1.596 3.804
LogisticRegression 0.860 0.737 0.854 0.691 0.857 0.823 0.928 0.657 0.825 1.880 3.520
NaiveBayes 0.754 0.713 0.835 0.665 0.834 0.798 0.906 0.611 0.802 1.890 3.510
DecisionTree 0.694 0.604 0.689 0.600 0.733 0.740 0.787 0.424 0.705 2.086 3.314
--- Aggregated Confusion Matrix Sums (Standard) ---
model TN_sum FP_sum FN_sum TP_sum
SVM 1394 106 267 933
RandomForest 1261 239 379 821
KNN 1400 100 502 698
LogisticRegression 1392 108 368 832
NaiveBayes 1359 141 396 804
DecisionTree 1181 319 476 724
--- Specificity / Sensitivity (Standard) ---
model Specificity Sensitivity
SVM 0.929 0.777
RandomForest 0.841 0.684
KNN 0.933 0.582
LogisticRegression 0.928 0.693
NaiveBayes 0.906 0.670
DecisionTree 0.787 0.603