ROADHEADER PERFORMANCE PREDICTION USING MACHINE LEARNING METHODS CASE STUDY: SAN MANUEL MINE, ARIZONA. NAZARBAYEV UNIVERSITY SCHOOL OF MINING AND GEOSCIENCES.

Abstract

This thesis is dedicated to the development and evaluation of predictive models for the performance of roadheader machines (ICR) using machine learning algorithms under conditions of limited data. The scarcity of available datasets is primarily due to high collection costs, issues of commercial confidentiality, and heterogeneous geological conditions, which significantly complicates the application of traditional prediction models. To address this challenge, the study employs data synthesis techniques that expand the training set by generating artificial observations through the addition of Gaussian noise, as well as alternative approaches based on Ridge Regression and Random Forest methods. A comparative analysis of various models is conducted, including linear methods (Ridge, Lasso, ElasticNet), ensemble algorithms (Random Forest, Gradient Boosting, Extra Trees), and nonlinear approaches (SVR,MLP). The results demonstrate that ensemble methods achieve the highest prediction accuracy, as evidenced by high R² values and low MSE values, even when using synthetically expanded datasets. However, while data synthesis improves model performance, it does not fully replace real-world observations, necessitating further validation of the developed models under practical conditions. The findings hold practical significance for optimizing planning processes and economic evaluations in the mining and construction industries, and they point to the promising prospects of integrating data synthesis techniques with real-time monitoring systems to enhance the robustness and interpretability of predictive models.