LCGA核對

Lasso Ranking

psqi_lcga_group_2

[模式] 排除模式

[資料] 來源=isi_raw_data  目標=psqi_lcga_group_2  樣本=149  特徵=61
  已排除群組：['ACS', 'EEG', 'ISI', 'PSQI']
  總缺值比例：6.91%

[CV結果]（分數 = neg_log_loss，越大越好 → log_loss 越小）
  lambda_min：C = 0.212095  |  lambda = 4.71487
  lambda_1SE：C = 0.0537848  |  lambda = 18.5926

[選入變項（1SE）] 以 |係數| 排序（前 30）
            coef  abs_coef
BDI_T1  0.315354  0.315354

[對照] lambda_min 非零變項數：14，lambda_1SE 非零變項數：1

[選入變項（lambda_min）] 以 |係數| 排序（前 30）
                          coef  abs_coef
BDI_T1                0.544447  0.544447
HRV_PNN50            -0.401001  0.401001
HRV_LF                0.324804  0.324804
EF_MOTIVATION         0.213037  0.213037
BAI_T1                0.110422  0.110422
HRV_RMSSD_MS         -0.082941  0.082941
WCST_PCT_PERS_RESP_T -0.080662  0.080662
IGT_NET_3            -0.077134  0.077134
HRV_NN50             -0.051903  0.051903
WM_SCALE_SCORE       -0.024061  0.024061
CPT_VAR_T            -0.020114  0.020114
HRV_RESP_RATE        -0.019821  0.019821
ERQ_ES                0.019131  0.019131
EF_PLANNING           0.004117  0.004117

[Top 10（路徑峰值）] 不綁定單一 C
        HRV_RMSSD_MS
WCST_PCT_PERS_RESP_T
         HRV_SDNN_MS
           IGT_NET_3
              BAI_T1
           HRV_PNN50
           CPT_PRS_T
    WCST_PERS_RESP_T
              HRV_LF
         CPT_HRTSD_T

psqi_lcga_group_3

[模式] 排除模式

[資料] 來源=isi_raw_data  目標=psqi_lcga_group_3  樣本=150  特徵=61
  已排除群組：['ACS', 'EEG', 'ISI', 'PSQI']
  總缺值比例：6.86%

[CV結果]（分數 = neg_log_loss，越大越好 → log_loss 越小）
  lambda_min：C = 0.104468  |  lambda = 9.57229
  lambda_1SE：C = 0.0562187  |  lambda = 17.7877

[選入變項（1SE）] 以 |係數| 排序（前 30）
                    coef  abs_coef
BDI_T1         -0.255188  0.255188
EF_ENV_MONITOR -0.106828  0.106828

[對照] lambda_min 非零變項數：6，lambda_1SE 非零變項數：2

[選入變項（lambda_min）] 以 |係數| 排序（前 30）
                    coef  abs_coef
BDI_T1         -0.393083  0.393083
EF_ENV_MONITOR -0.269334  0.269334
HRV_LF         -0.061635  0.061635
IGT_DECK_D      0.036825  0.036825
HRV_LF_HF      -0.029828  0.029828
BAI_T1         -0.025573  0.025573

[Top 10（路徑峰值）] 不綁定單一 C
       EF_ENV_MONITOR
WCST_PCT_CONCEPTUAL_T
               HRV_LF
               ERQ_CR
          EF_PLANNING
          HRV_SDNN_MS
  WCST_TOTAL_ERRORS_T
               BDI_T1
             HRV_NN50
               BAI_T1

---

BAI_T1, BDI_T1基準

psqi_lcga_group_2

[資料] 來源=isi_raw_data  目標=psqi_lcga_group_2  列數=149  特徵數=2
[特徵] 使用欄位（前 15）：['BDI_T1', 'BAI_T1']
[CV] Stratified 5-fold, seed=42  |  class_weight=balanced
[聚合] K-fold = weighted | LOSO = weighted

[Leakage check] Class balance
                   count  percent%
psqi_lcga_group_2                 
0                    111      74.5
1                     38      25.5

[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
LogisticRegression 0.713       0.512     0.458    0.579       0.802     0.842    0.766 0.322     0.718       9.600      20.200
        NaiveBayes 0.706       0.444     0.560    0.368       0.851     0.806    0.901 0.314     0.765       5.000      24.800
               SVM 0.676       0.523     0.460    0.605       0.800     0.848    0.757 0.334     0.718      10.000      19.800
               KNN 0.666       0.467     0.636    0.368       0.866     0.811    0.928 0.364     0.785       4.400      25.400
      DecisionTree 0.635       0.458     0.422    0.500       0.791     0.817    0.766 0.252     0.698       9.000      20.800
               MLP 0.593       0.421     0.421    0.421       0.802     0.802    0.802 0.223     0.705       7.600      22.200
      RandomForest 0.593       0.423     0.455    0.395       0.819     0.802    0.838 0.244     0.725       6.600      23.200
           XGBoost 0.555       0.409     0.360    0.474       0.752     0.798    0.712 0.171     0.651      10.000      19.800

--- Aggregated Confusion Matrix Sums (across all folds' test parts) ---
             model  TN_sum  FP_sum  FN_sum  TP_sum
LogisticRegression      85      26      16      22
        NaiveBayes     100      11      24      14
               SVM      84      27      15      23
               KNN     103       8      24      14
      DecisionTree      85      26      19      19
               MLP      89      22      22      16
      RandomForest      93      18      23      15
           XGBoost      79      32      20      18

psqi_lcga_group_3

[資料] 來源=isi_raw_data  目標=psqi_lcga_group_3  列數=150  特徵數=2
[特徵] 使用欄位（前 15）：['BDI_T1', 'BAI_T1']
[CV] Stratified 5-fold, seed=42  |  class_weight=balanced
[聚合] K-fold = weighted | LOSO = weighted

[Leakage check] Class balance
                   count  percent%
psqi_lcga_group_3                 
0                     84      56.0
1                      8       5.3
2                     58      38.7

[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 0.686       0.587     0.604    0.620       0.477     0.520    0.476 0.255     0.620       1.200      22.600
               SVM 0.650       0.569     0.600    0.553       0.439     0.448    0.482 0.225     0.553       5.000      16.600
LogisticRegression 0.648       0.564     0.622    0.560       0.436     0.468    0.513 0.252     0.560       6.200      18.200
      RandomForest 0.638       0.579     0.565    0.593       0.397     0.390    0.405 0.206     0.593       0.400      18.400
               MLP 0.630       0.575     0.571    0.580       0.436     0.442    0.433 0.200     0.580       1.200      17.600
           XGBoost 0.612       0.597     0.596    0.613       0.453     0.474    0.447 0.244     0.613       1.000      20.200
      DecisionTree 0.600       0.568     0.570    0.567       0.456     0.452    0.463 0.194     0.567       2.000      17.000
               KNN 0.588       0.555     0.554    0.573       0.417     0.431    0.416 0.166     0.573       1.200      20.400

---