Authors: Yu.N. Linnik, V.Yu. Linnik

Title of the article: Determination of coal quality characteristics using artificial intelligence algorithm

Year: 2026, Issue: 3, Pages: 195-204

Branch of knowledge: 2.8.9. Mineral processing

Index UDK: 519.683

DOI: 10.26730/1999-4125-2026-3-195-204

Abstract: The primary objective of coal producers in today's economic environment is not merely to increase gross output volumes, but to ensure the comprehensive supply of fuel that strictly meets specified quality parameters, while simultaneously minimizing operating and capital expenditures across the entire mining cycle. Given the volatility of market prices and the tightening of environmental standards for combusted fuel, accurate prediction of key coal quality attributes—such as ash content, volatile matter yield, calorific value, and sulfur content—becomes paramount. This enables the optimization of beneficiation, blending, and logistics processes, thereby maximizing the energy utilization potential of every ton of extracted raw material. Consequently, the aim of this paper is not only to provide a superficial review, but to conduct an in-depth practical investigation and comparative evaluation of a set of common machine learning and artificial intelligence algorithms applied to a real-world production task—forecasting coal quality. The empirical foundation of the study is built upon an extensive dataset of laboratory analyses, collected and verified over a five-year period (2015–2020). The sample comprises 33,256 representative coal specimens sourced from various seams of the Kuznetsk coal basin, ensuring high statistical significance of the findings. In the experimental modelling, four fundamentally different approaches were analyzed: the C4.5 decision tree, the IBk nearest neighbor method, the naïve Bayes classifier, and the multilayer perceptron (MLP). The core idea of the work was to identify the most suitable model: each algorithm underwent careful probabilistic output calibration to obtain unbiased predictions, the influence of each input parameter on the final result was assessed, and a final ranking of the methods was performed based on accuracy and robustness criteria. Based on the comprehensive evaluation, the MLP was recognized as the most effective tool for this domain, with its topology—featuring optimally tuned numbers of neurons in the input, hidden, and output layers—demonstrating the best trade-off between training speed and generalization capability.

Key words: artificial intelligence algorithms intelligent data analysis classification predictive model coal quality

Receiving date: 17.02.2026

Approval date: 15.06.2026

Publication date: 27.08.2026

CITE DOWNLOAD