Unmasking effects of feature selection and SMOTE-Tomek in tree-based random forest for scorch occurrence detection

Abstract

Scorch occurrence during the production of flexible polyurethane foam has been a menace that consistently, jeopardize a foam’s integrity and resilience. It leads to foam suppression and compactness integrity failure due to scorch. There is always the increased likelihood of scorching, and makes crucial the utilization of methods that seek to avert it. Studies predict that the formation of foam constituent processes via optimization using machine learning have adequately trained models to effectively identify scorch occurrence during the profiling in the polyurethane foam production. Our study utilizes the random forest (RF) ensemble with feature selection (FS) and data balancing technique to identify production predictors. Study yields accuracy of 0.9998 with F1-score of 0.9819. Model yields 2-distinct cases for (non)-occurrence of scorch respectively, and the ensemble demonstrates that it can effectively and efficiently predict the occurrence of scorch in the production of flexible polyurethane foam manufacturing process.