ISI預測項目:BAI_T1, BDI_T1,HRV

  • 篩選方式

    • ISI_T1, ISI_T2, ISI_T3 都>=9 (標註為1)
    • ISI_T1, ISI_T2, ISI_T3 都<9 (標註為0)
  • 樣本數

    • 篩選條件(HRV_SDNN_MS, HRV_LF_HF, HRV_LF, BDI_T1, BAI_T1)不缺
      • 3TP = 0 (健康組):35 人 (佔 68.6%)
      • 3TP = 1 (失眠組):16 人 (佔 31.4%)

原始預測結果:

  • 只考慮BAI_T1, BDI_T1

HRV預測相關性:

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_transformer_abs  目標=3TP  樣本=80  特徵=12
  已排除群組:['ACS', 'CDRISC', 'CPT', 'EEF', 'EF', 'ERQ', 'IGT', 'WCST']
  總缺值比例:36.25%

[CV結果](分數 = neg_log_loss,越大越好 → log_loss 越小)
  lambda_min:C = 0.136775  |  lambda = 7.31127
  lambda_1SE:C = 0.0316228  |  lambda = 31.6228

[選入變項(1SE)] 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_1SE 非零變項數:0

[Top 10(路徑峰值)] 不綁定單一 C
        HRV_PNN50
         HRV_NN50
        HRV_POWER
HRV_EKG_HR_MAXMIN
    HRV_RESP_RATE
     HRV_RMSSD_MS
           HRV_LF
       HRV_EKG_HR
          HRV_VLF
        HRV_LF_HF

进程已结束,退出代码为 0

Elastic Net Ranking:

/Users/yuchi/PycharmProjects/PsyMl_Data/.venv/bin/python /Users/yuchi/PycharmProjects/PsyMl_ISI/ML/tools/elastic_net_ranking.py 
[模式] 允許模式

[資料] 來源=isi_raw_data_transformer_abs  目標=3TP  樣本=80  特徵=12
  已排除群組:['ACS', 'CDRISC', 'CPT', 'EEF', 'EF', 'ERQ', 'IGT', 'WCST']
  總缺值比例:36.25%
  使用模型:Elastic Net (l1_ratio=0.4)

[CV結果](分數 = neg_log_loss,越大越好 → log_loss 越小)
  lambda_min:C = 0.0840388  |  lambda = 11.8993
  lambda_1SE:C = 0.0316228  |  lambda = 31.6228

[選入變項(1SE)] 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:1,lambda_1SE 非零變項數:0

[選入變項(lambda_min)] 以 |係數| 排序(前 30)
            coef  abs_coef
HRV_LF  0.008315  0.008315

[Top 10(路徑峰值)] 不綁定單一 C
        HRV_PNN50
         HRV_NN50
        HRV_POWER
HRV_EKG_HR_MAXMIN
    HRV_RESP_RATE
     HRV_RMSSD_MS
       HRV_EKG_HR
           HRV_LF
        HRV_LF_HF
      HRV_SDNN_MS

過去加上HRV特徵:

  • 使用欄位:HRV_SDNN_MS, HRV_LF_HF, HRV_LF, BDI_T1, BAI_T1

  • 使用欄位:HRV_LF, BDI_T1, BAI_T1

CV拆解 test_size=0.2

LOSO

SHAP貢獻度