Assessment of the environmental impact of hard coal mining waste disposal sites using machine learning algorithms, with an indication of important features influencing the selection of learning algorithms

Abstract

This paper presents the potential application of supervised machine learning algorithms to assess the environmental impact of mine disposal sites. Algorithms available for Python from the scikit-learn and pandas libraries were applied to a group of sites representing mine waste dumps used for the disposal of hard coal mining waste. Each disposal site was described with 11 attributes (site characteristics, waste characteristics, groundwater, surface water, air, soil, atmospheric factors, geology, geohazards, nature, and human environment) and 73 features (categorical, numerical, and descriptive) detailing the sites’ environmental impact. As a result of applying the learning process to training data and verifying it on test data, prediction results of at least 80% were obtained for all algorithms tested. The results indicate that the best algorithm for determining the environmental impact of the waste dumps would be the BernouliNB algorithm (86% prediction accuracy), followed by the RidgeClassifier algorithm (87% prediction accuracy), with the currently available training dataset.  Potential extension of the dataset could improve the results of the MLPClassifier, Support Vector Machine, and LogisticRegression algorithms.