Fig. 4: Sparse ML model development via input feature pruning for clinker phase.
From: Industrial-scale prediction of cement clinker phases using machine learning
a MAPE of optimal machine learning models (Neural Network for alite, Gaussian Process Regression for belite, Support Vector Regression for ferrite) across 15 combinations of input features: process parameters (PP), kiln feed (KF), hot meal (HM), and clinker oxides (CO). Values in parentheses represent MAPE (%) for alite (green), ferrite (blue), and belite (red) predictions. b Alite prediction using NN, (c) Belite prediction using GPR, and (d) Ferrite prediction using SVR, utilizing 10 controllable PP and KF compositions. The top-right inset shows each phase’s best-performing model--NN for alite, GPR for belite, and SVR for ferrite--when trained with all 59 input features. Black circles represent the mean model-prediction, while red error bars indicate model uncertainty ( ± 3σ). The bottom-left inset presents a comparative analysis of MAPE for the sparse model against the best-performing model, as well as the Bogue and clinker equations developed in this study.
