Skip to main content

Table 2 Cross-validated, TT-corrected performances of SES and LASSO-type methods on the four scenarios

From: Feature selection for high-dimensional temporal data

Temporal-longitudinal scenario Temporal-distinct scenario
  MSPE Selected vars   MSPE Selected vars
Dataset SESglmm glmmLasso SESglmm glmmLasso Dataset SES LASSO SES LASSO
GDS5088 0.081 (0.026) 0.160 (0.042) 5.25 (0.85) 5.15 (8.65) GDS3859 0.068 (0.006) 0.019 (0.002) 3.5 (0.51) 11.81 (4.66)
GDS4395 0.104 (0.041) 0.640 (0.568) 5.37 (0.56) 12.35 (13.61) GDS972 0.022 (0.000) 0.001 (0.000) 5.83 (0.92) 22.2 (9.85)
GDS4822 0.115 (0.484) 0.765 (0.436) 4.75 (0.85) 3.16 (5.77) GDS947 0.056 (0.000) 0.054 (0.026) 5.92 (0.65) 12.40 (5.40)
GDS3326 0.135 (0.021) 0.234 (0.139) 5.42 (0.78) 2.42 (7.45) GDS964 0.033 (0.000) 0.003 (0.000) 5.73 (0.69) 25.69 (11.86)
GDS3181 0.971 (0.484) 0.684 (0.257) 4.17 (0.87) 0.35 (2.15) GDS2688 0.184 (0.006) 0.005 (0.001) 5.79 (1.06) 20.64 (10.93)
GDS4258 0.234 (0.096) 9.882 (4.518) 3.83 (0.51) 1.48 (4.06) GDS2135 0.053 (0.002) 0.014 (0.003) 3.80 (0.76) 10 (5.72)
GDS3432 0.357 (0.017) 2.283 (1.572) 1.67 (3.51) 0.08 (0.55) Av. diff. 0.053 a -12.03 a
GDS3915 0.059 (0.002) 0.150 (0.055) 5.12 (0.80) 1.66 (4.62)  
Av. diff. -1.59 b 1.12 b  
Static-distinct scenario Static-longitudinal scenario
  PCC Selected vars   PCC Selected vars
Dataset SES LASSO SES LASSO Dataset SES GLASSO SES GLASSO
GDS4319 0.873 (0.000) 0.995 (0.000) 2.1 (0.31) 8 (0.00) GDS4146 1.000 (0.000) 0.858 (0.142) 1.00 (0.00) 0.42 (1.38)
GDS3924 0.729 (0.000) 0.528 (0.104) 2.75 (0.44) 53.56 (28.55) GDS4518 0.750 (0.000) 0.417 (0.333) 1.75 (0.44) 3.04 (2.15)
GDS3184 0.556 (0.067) 0.578 (0.111) 3.00 (0.00) 10.62 (5.16) GDS4820 0.500 (0.000) 0.667 (0.167) 2.00 (0.00) 5.14 (3.19)
GDS3145 0.953 (0.000) 0.594 (0.125) 1.5 (0.88) 0.6 (0.55) GDS1840 0.625 (0.000) 0.500 (0.250) 1.5 (0.51) 2.67 (2.03)
GDS2882 0.800 (0.000) 0.750 (0.000) 1.5 (0.88) 0.25 (0.50) Av. diff. 0.108 -1.23
GDS2851 0.722 (0.000) 0.694 (0.000) 2.25 (0.44) 0.75 (0.50)  
GDS1784 0.861 (0.000) 0.694 (0.000) 1.75 (0.85) 0.5 (0.58)  
GDS2456 1.000 (0.000) 0.739 (0.000) 1.2 (0.41) 0.44 (0.53)  
Av. diff. 0.115 b -6.52 b  
  1. For each dataset, performances are reported as average (st.d.). Zero standard deviations are caused by numerical rounding. For Temporal-longitudinal and Temporal-distinct scenario’s performance are computed as Mean Squared Prediction Error (MSPE, lower values indicate better performances) and number of selected variables, while for the other scenarios the Percentage of Corrected Classification (PCC, the higher the better) is used instead of MSPE. The bold numbers indicate better performance; average differences over all datasets are reported for each scenario. Symbols a and b denote average differences that are statistically significant at 0.01 and 0.05, respectively. In terms of predictive performances, SES is always on par or better than LASSO type algorithms in all scenarios except for the Temporal-distinct