Prediction of Various Sizes of Particles in Deep Opencast Copper Mine Using Recurrent Neural Network: A Machine Learning Approach

Human health, socioeconomics, agriculture, and the environment have been impacted by the total concentration of particulate matter (PM). The main aim of the proposed work was to predict the concentration of particles at various depths and heights using supervised machine learning (ML) algorithms such as support vector regression (SVR), random forest regression, XGBoost regression, and deep learning (DL) algorithms such as recurrent neural network (RNN) and long short-term memory (LSTM) to assess the behavior of various sizes of particles and their movement from the source to surface in deep opencast mine. In this regard, the dispersion of multiple particle sizes is significant from a health point of view, as longer retention time increases exposure to dust and adversely impacts the health of mineworkers. Total 20 factors (15 different sizes of particles, four meteorological parameters, and depth/height inside the mine) were applied to different algorithms. In ML methods, it has been observed that SVR performed better than other regression models with a total mean absolute error of 9.58. The RNN–LSTM predicts the PM values for the various range sizes with better accuracy. Predictions ensure rapid response to the pattern and precision to the actual values. The results indicated that DL approaches have significant applicability to forecast the concentration profile of particles to reduce PM exposure to human health. The study concluded that air quality decision/policymakers and experts could apply DL approaches to estimate the spatiotemporal concentration of particles inside the mines and plan to reduce workers’ exposure to it. Empirical equations have been proposed showing the relationship between particles in 15 different sizes and depth.