With the fast growth of aquatic data, machine learning is essential for data analysis, categorization, and prediction. Data-driven models using machine learning may effectively handle complicated nonlinear problems in water research, unlike conventional approaches. Machine learning models and findings have been used to build, monitor, simulate, evaluate, and optimize water treatment and management systems in water environment research. Machine learning may also enhance water quality, pollution control, and watershed ecosystem security. This chapter discusses how ML approaches were used to assess water quality in surface, ground, drinking, sewage, and ocean. The authors also suggest potential machine learning applications in aquatic situations.
Top1. Introduction
Wastewater carrying toxins from fast economic growth threatens natural water ecology. So, several water pollution management methods evolved. Water quality analysis and assessment have greatly enhanced water pollution control efficiency. The multivariate statistical approach, fuzzy inference, and water quality index (WQI) are among the various methods used to monitor and measure water quality globally. While many water quality metrics may be traced in accordance with regulations, the final findings may vary according to parameter selection. Taking into account that all water quality metrics is impractical due to cost, technical difficulty, and inability to account for variability. Recently, developments in machine learning have led academics to anticipate that huge volumes of data may be achieved and assessed to accomplish complex and large-scale water quality monitoring needs.
ML algorithms are used in artificial intelligence to examine data and find patterns to forecast future information. With its accuracy, flexibility, and extensibility, machine learning has become a popular data analysis and processing tool in several fields. Machine learning simplifies the finding of underlying mechanisms for complex nonlinear relational data. Recently, ML showed a huge potential as a tool in ecological science and engineering due to its versatility. In spite of the difficulty of ML for water quality measurement and assessment, much precise results are predicted.
Complex water kinds include drinking, wastewater, and groundwater, surface, marine, and fresh. These water kinds have varied qualities, making quality study difficult. Previous research suggests that machine learning may effectively handle these difficulties. In this study, we address the pros and cons of typical ML approaches and their implementations and performance in surface water, groundwater, drinking water, wastewater, and ocean water (Fig. 1).
ML is commonly utilized to find insights or have predictions from vast data from many contexts. Prior to using ML, data collecting, algorithm selection, model training, and validation are needed. Among these methods, selecting algorithm is the key aspect.
Figure 1. Water systems employ machine learning extensively. WWTP, wastewater treatment facility
Machine learning has two primary classes: supervised and unsupervised. Labels in datasets distinguish these two kinds. Supervised learning predicts from labeled training datasets. Input and anticipated output values are included in each training instance. Supervised learning algorithms discover input-output correlations and create a predictive model to estimate the outcome from the I/P data. Supervised learning methods, such as LR, ANN, decision trees, SVM, Naive Bayes, KNN, and random forests are designed for data classification and regression.
In contrast, unsupervised learning handles data generally without labels, addressing pattern recognition problems using unlabeled training datasets. Unsupervised learning classifies training data depending on features, primarily via dimensionality reduction and grouping. But, the quantity and significance of categories are unknown. Thus, unsupervised learning is often employed for classification and association mining. PCA, K-means, and other unsupervised machine learning methods are popular. Reinforcement learning, which allows machines to generalize and solve unlearned problem is a different kind of ML method. In contrast to the other 2 ML classes, it is sometimes used in the aquatic aspect.