Fig. 3: Summary of random forest classifier (RFC) performance for identifying non-negative matrix factorization (NMF) cluster in ALS patients.
From: Machine learning predicts distinct biotypes of amyotrophic lateral sclerosis
One verse rest (OvR) RFC results from cortex (CTX) and spinal cord (SC) analyses. Confusion matrices, corresponding to predicted NMF cluster (columns) and actual cluster (rows) (A); ROC curves, true vs false positive rates plotted along with the AUC with different style lines for the four NMF classifiers along with a micro-average of the four (dotted blue line) and a dotted black line for chance-based performance (B); selected model performance metrics (C); and a bar plot of the top 10 most important features along with their permutation importance scores (D) for RFC-based models in the CTX (left) and SC (right). Under the NMF ID, the pathophysiological relevance of the cluster is indicated by a R (neuronal regeneration), SD (synaptic dysfunction), D (neuronal degeneration), or C (control).
