Skip to main content

Table 3 Overview of the generalization performance (balanced accuracy) of each model when trained using toy data

From: LANDMark: an ensemble approach to the supervised selection of biomarkers in high-throughput sequencing data

 

LANDMark (oracle)

LANDMark (no oracle)

Extra trees

LinearSVC

Logistic regression

Random forest

Ridge regression

SGD (MH)

SGD (SH)

Two spirals

0.935 ± 0.058 (3)

0.939 ± 0.047 (2)

0.949 ± 0.043 (1)

0.445 ± 0.080 (9)

0.491 ± 0.077 (6)

0.866 ± 0.062 (4)

0.495 ± 0.064 (5)

0.487 ± 0.104 (7)

0.457 ± 0.106 (8)

Concentric circles

0.844 ± 0.046 (1)

0.837 ± 0.042 (2)

0.781 ± 0.049 (3)

0.364 ± 0.116 (7)

0.428 ± 0.058 (5)

0.802 ± 0.064 (4)

0.425 ± 0.060 (6)

0.305 ± 0.079 (9)

0.348 ± 0.082 (8)

Parkinson’s disease

0.718 ± 0.082 (2)

0.719 ± 0.093 (1)

0.677 ± 0.076 (5)

0.679 ± 0.086 (4)

0.693 ± 0.081 (3)

0.667 ± 0.085 (6)

0.635 ± 0.080 (7)

0.554 ± 0.099 (9)

0.564 ± 0.099 (8)

Iris

0.926 ± 0.038 (3)

0.918 ± 0.037 (4)

0.928 ± 0.037 (1)

0.858 ± 0.076 (7)

0.890 ± 0.083 (5)

0.926 ± 0.035 (2)

0.746 ± 0.041 (9)

0.874 ± 0.068 (6)

0.796 ± 0.129 (8)

Breast cancer

0.945 ± 0.014 (1)

0.940 ± 0.017 (3)

0.929 ± 0.023 (7)

0.942 ± 0.014 (2)

0.938 ± 0.021 (4)

0.919 ± 0.024 (8)

0.900 ± 0.024 (9)

0.929 ± 0.021 (6)

0.931 ± 0.027 (5)

Wine

0.974 ± 0.032 (3)

0.973 ± 0.031 (5)

0.976 ± 0.032 (2)

0.967 ± 0.032 (7)

0.974 ± 0.030 (4)

0.971 ± 0.031 (6)

0.977 ± 0.027 (1)

0.965 ± 0.039 (8)

0.946 ± 0.046 (9)

Seeds

0.902 ± 0.022 (3)

0.905 ± 0.021 (2)

0.877 ± 0.039 (7)

0.890 ± 0.031 (4)

0.887 ± 0.022 (5)

0.849 ± 0.053 (9)

0.931 ± 0.023 (1)

0.874 ± 0.043 (6)

0.851 ± 0.074 (8)

Heart failure

0.622 ± 0.057 (1)

0.621 ± 0.064 (2)

0.557 ± 0.069 (6)

0.526 ± 0.111 (8)

0.588 ± 0.085 (4)

0.613 ± 0.061 (3)

0.574 ± 0.068 (5)

0.390 ± 0.131 (9)

0.546 ± 0.117 (7)

Cancer coimbra

0.714 ± 0.096 (1)

0.700 ± 0.107 (2)

0.690 ± 0.099 (3)

0.644 ± 0.124 (8)

0.580 ± 0.127 (4)

0.656 ± 0.114 (6)

0.669 ± 0.135 (5)

0.655 ± 0.121 (7)

0.623 ± 0.123 (9)

Raisin

0.839 ± 0.026 (3)

0.839 ± 0.027 (2)

0.815 ± 0.031 (8)

0.827 ± 0.049 (5)

0.840 ± 0.034 (1)

0.825 ± 0.032 (6)

0.832 ± 0.034 (4)

0.819 ± 0.104 (7)

0.813 ± 0.035 (9)

Two moons

0.968 ± 0.057 (2)

0.978 ± 0.041 (1)

0.968 ± 0.052 (3)

0.761 ± 0.169 (9)

0.803 ± 0.086 (6)

0.934 ± 0.081 (4)

0.817 ± 0.075 (5)

0.774 ± 0.090 (8)

0.778 ± 0.146 (7)

HMP

0.953 ± 0.022 (3)

0.948 ± 0.019 (7)

0.936 ± 0.025 (8)

0.955 ± 0.021 (1)

0.950 ± 0.018 (4)

0.923 ± 0.033 (9)

0.949 ± 0.027 (5)

0.954 ± 0.022 (2)

0.949 ± 0.026 (6)

Average Ranking

2.17

2.75

4.5

5.92 *

4.25

5.58

5.17

7 **

7.67 **

  1. The mean, sample standard deviation, and rank for each model is reported. The best performing models are highlighted in bold text. Asterisks indicate a statistically significant (p ≤ 0.05) difference between models. A single asterisk indicates a difference in favor of LANDMark (Oracle) while a double asterisk indicates a difference in favor of either LANDMark model