Evaluation of Machine Learning and Genetic Algorithms for Water Quality Prediction

The state of the water is referred to as its quality. Both natural processes and urbanization are lowering the quality of the water. Drinking such water is dangerous. Water must therefore be evaluated before being used. Sufficient data for training and assessment are necessary for an efficient model of water quality assessment. To improve the precision and effectiveness of the water quality model, they ought to record the statistical characteristics or trends of the dataset's water quality indicators. The dynamic nature of water quality parameter values poses challenges in identifying pertinent elements and patterns. Another difficulty is that different sample sites have different water quality requirements, and different consortiums have different standards for evaluating water quality. In order to find pertinent features for the categorization and prediction of water quality, this work use genetic algorithms in conjunction with empirical research on potential machine and deep learning models. Finding the best water quality prediction models across a range of datasets is the objective of this research. After the features were identified, evaluated the datasets using machine learning and deep learning approaches such as Random Forest (RF), Logistic Regression (LR), Ada Bost (AB), Extreme Gradient Bossing Machine (XGB), and Deep Neural Net. The empirical analysis shows that the LR model is yields superior (98.61%) accuracy followed by the Random Forest model (with accuracy 98.10%) on Drinking Water Final_Dataset. The DNN model outperforms the other models with 96.18% accuracy on Aquaattributes Dataset. The RF model records considerable accuracy 67.55% than the other model over the Water_potability dataset. The XGB model performed well on the Water quality analysis dataset. From the empirical analysis it is noticed that the accuracy should increase if pertinent features are trained into the model.