Gene targeting in amyotrophic lateral sclerosis using causality-based feature selection and machine learning

Table 3 Comparison of model performance trained on different numbers of selected genes

	473 genes	44 genes	Full dimensionality
XGBoost classifiers
Pooled cerebellum and frontal cortex samples	72.33%	67.02%	44.25%
Cerebellum samples	70.37%	85.18%	48.14%
Frontal cortex samples	85.18%	81.48%	51.85%
Random Forest classifiers
Pooled cerebellum and frontal cortex samples	74.22%	79.75%	46.96%
Cerebellum samples	37.03%	40.74%	25.92%
Frontal cortex samples	29.62%	40.74%	25.92%

Accuracy achieved by the XGBoost and Random Forest classifiers for the classification of the hold-out datasets when they were trained on (a) the expression values of the 473 geness, (b) the expression values of the 44 genes, (c) full dimensionality (expression values of 23,188 genes). The analysis was performed for (i) pooled cerebellum and frontal cortex samples, (ii) only for the cerebellum samples, as well as (c) only for the frontal cortex samples. The most accurate classification cases are highlighted in italics

ISSN: 1528-3658