Machine learning-based water requirement forecast and automated water distribution control system

Abstract

Wastage of water is a burning topic in the world. Different countries worldwide are facing the issue of the lack of fresh water, and the problem is increasing daily. This paper aims to design a system that will predict the amount of water needed by a family or in a locality depending on family members, region, temperature, season, occupation, location, and religion. It can also be possible to forecast the water demand in a locality, area, or country depending on these factors. Previous works in this field focus on something other than these factors and the distribution system mentioned in this paper. Different machine learning models will predict the amount of water required by a family or a locality based on these factors. Then, water will be supplied using these expected values so that each family or community in a locality receives the desired amount of water. The practical circuit uses an Arduino microcontroller, water flow meter, solenoids, etc. Water distribution is automatically controlled by the water flow meter and solenoid, and no family or community in a locality will receive more water than the predicted values per day. So it will reduce water wastage, and everybody will use it according to their daily needs. Different machine learning models were used in this proposed design to compare the performance of the models for this task. Linear, Ridge, Lasso, ElasticNet, Decision Tree, Random Forest, XGBoost (Extreme Gradient Boosting), KNN (K-Nearest Neighbors), SVR (Support Vector Regression), MLP (Multilayer Perceptron), LightGBM (Light Gradient-Boosting Machine), CatBoost, Deep Neural Network have been used. Different model's performances have been analyzed. The analyzing factors are model training time, model prediction time, Robustness to outliers, and scalability. All these performances were analyzed to determine which model is best for this work. So, the Decision Tree and LightGBM models are the best based on comparing all the models for this task. Keywords: Factors Influencing Water Consumption, Different Machine Learning Models Comparison, Water Demand Prediction and Forecast, Precise Water Distribution, Reducing Water Wastage.