Abstract
Regular Water quality (WQ) monitoring is vital for environmental and public health protection, especially in tailing dams. Failures of tailing dams can cause catastrophic environmental damage and loss of life, with a greater default rate than traditional water retention dams. Regular maintenance and monitoring ensure stability and safety.
Effluent treatment plants for tailings dams are essential for managing and treating the water and waste materials generated from mining operations. These plants help ensure that the effluent released into the environment meets regulatory standards and minimizes environmental impact. Constantly tracking WQ parameters is vital for verifying the handling process's efficiency. Automated systems and sensors can facilitate the monitoring of parameters like pH, turbidity, and contamination level in real time.
Machine learning (ML) is a vital tool for managing large datasets since it enables software to predict outcomes without the need for specialized programming. ML uses mathematical principles to develop algorithms for precise predictions, such as classifying water in effluent treatment plants. Over time, big data computing technology has evolved, allowing algorithms automatically employing complex calculations to large datasets, resulting in faster results.
ML holds significant potential for improving WQ prediction and classification. This study assesses various AI algorithms to manage long-term WQ dataset and develop a reliable technique for the most accurate WQ forecasts. Including the evaluation of the Tree, SVM, Naïve Bayes, logistic regression kernel, KNN, and SVM kernel. These algorithms were trained using the classification and regression learner app. This study involves various metrics to appraise the model’s operation. The dataset was split into two sections: 30% was used to forecast the models' performance, and 70% was used to train and construct the models.
The study evaluates the effluent water quality dataset from a tailing dam in Zambia, using online the monitoring platform. This platform enables proactive asset management by triggering alarms when parameters exceed thresholds. A dataset consisting of 41,450 samples was collected and analyzed, measuring pH, TSD,
5
temperature, flow, and total suspended solids. The data was gathered between August 30, 2024, and October 31, 2024, using sensors data at the tailing dam in Zambia.
Furthermore, this research paper use the Econometrics toolbox in MATLAB to test the stationary on the datasets. The ARIMA models were used to test and train water quality parameters, with the differential transformative model being the most effective in tracking variability. The model's accuracy and suitability for monitoring water quality improved by nearly matching actual readings. The datasets had p-values under 0.05, indicating stationarity, declining the null hypothesis and indicating consistency throughout. The tailings dam dataset was trained using built-in regression software, including tree, SVM, kernel, ensemble, and Neural Network (NN). The NN was the most remarkable algorithm, with high-performance metrics: RMS 0.2441, MSE 0.059587, R-Squared 0.98571, and MAE 0.1722. When compared to other machine learning models, SVM approaches have demonstrated the best results, with 97.6% accuracy, 97.55% precision, 97.87 recall, and 97.65% F1 score.