Estimation of water consumption amounts using machine learning algorithms: The case study of Kocaeli

The efficient management and sustainable use of water resources is one of the most critical environmental and economic issues of our time. Water resource management is not just about meeting current needs, but also about ensuring access to water for future generations. However, climate change, extreme weather events, and increasing water demand are making this management more complex. Water management must respond to the needs of different sectors, such as agriculture, industry, and domestic use, which requires a complex planning process. Factors such as increasing population, urbanization, and climate change are making the management of water resources increasingly complex. In this context, accurate prediction of water consumption amounts is of great importance for the planning and management of water distribution systems. Accurate predictions are vital for the optimal use of water resources and the prevention of unnecessary waste. Determining potential losses in water distribution networks in advance and being able to respond quickly to sudden changes in water demand are among the elements that increase the sustainability of water management. In addition, water consumption predictions play an important role in planning infrastructure investments and making long-term strategic decisions. This study aims to estimate water consumption amounts using various artificial intelligence techniques, with a primary focus on Artificial Neural Networks (ANN). While other models like Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRU) were considered, the emphasis is on ANN due to its robustness and adaptability to complex data structures. The research utilizes data obtained from the Kocaeli Water and Sewerage Administration General Directorate. This dataset provides a comprehensive view of regional water consumption patterns. The study analyzes the monthly water consumption data of 5000 subscribers, consisting of 3447 residential, 1422 commercial, and 131 public subscribers, between January 2019 and August 2022. Additionally, to improve the prediction performance for residential subscribers, daily water consumption data from 33 residential subscribers between September 3, 2019, and November 5, 2021, were used. Daily data were chosen to more accurately capture short-term variations in water consumption and enhance predictive accuracy. This dataset includes daily total water consumption values for each subscriber, as well as meteorological and socio-economic data. Subscribers with missing or inconsistent data were excluded from the analysis to ensure data quality. The data were pre-processed and prepared for use in daily consumption predictions, optimizing the models for higher accuracy. The dataset includes different subscriber types (residential, commercial, public) and various factors, including tariff types, activity types, weather data, seasonality, and COVID-19 pandemic data. The study analyzed the importance of these factors on water consumption and identified the most critical variables influencing water demand. Additionally, meteorological variables such as temperature, precipitation, and humidity were evaluated to understand seasonal changes in water consumption. The impact of the COVID-19 pandemic was particularly significant for residential subscribers, leading to an increase in water usage, while commercial and public subscribers experienced a decline. This shift in consumption patterns underscores the need for adaptive and responsive water management strategies. Various artificial intelligence techniques were applied in the study, with Artificial Neural Networks (ANN) being the primary method. Other machine-learning models, such as Support Vector Machines (SVM) and Random Forest (RF), were also explored for comparative analysis. While Deep Learning algorithms were considered, ANN was chosen for its balance between complexity and computational efficiency. Data augmentation techniques were employed to enable the model to adapt to different scenarios and provide more consistent predictions, particularly in the face of unforeseen events like the pandemic. These algorithms are capable of analyzing numerous variables simultaneously, and they excel in handling large datasets. The developed models were designed and trained to predict future water consumption amounts using factors affecting water consumption as inputs. Data preprocessing techniques such as filling in missing values, detecting and removing outliers, and data normalization were applied to ensure data quality and model accuracy. For model training, the dataset was divided into training, validation, and test sets, following standard machine learning practices. This approach allowed for rigorous evaluation and optimization of the models' performance. The models' performance was evaluated using various statistical metrics (R², MSE, RMSE, MAE). The results showed that the ANN model, in particular, was able to make high-accuracy predictions for commercial and official subscribers. Although the prediction performance for residential subscribers was relatively lower, meaningful results were still obtained, demonstrating the model's overall effectiveness. These metrics provide a quantitative basis for assessing the reliability and accuracy of the predictions. The study compared the performance of different artificial intelligence techniques and analyzed the strengths and weaknesses of each. Additionally, the importance levels of factors affecting water consumption were determined, and the effects of these factors on prediction performance were examined. Tariff types, activity types, seasonality, and the COVID-19 pandemic were found to have a significant impact on water consumption. It was found that tariff types, activity types, seasonality, and the COVID-19 pandemic had significant effects on water consumption. To increase the generalization ability of the models, cross-validation techniques were used, and hyperparameter optimization was performed. Feature importance analysis was conducted to enhance the interpretability of the models, allowing for the identification of which factors have the most significant impact on water consumption. This rigorous approach ensures that the models are robust and reliable for real-world applications. This research demonstrates the potential of artificial intelligence techniques, particularly ANN, in predicting water consumption and provides a valuable tool for water resource management. The findings can be used for planning water distribution systems, developing demand management strategies, and promoting sustainable water use. The results of the study also offer new perspectives for future research and providea foundation for further development of artificial intelligence applications in water consumption estimation. Additionally, the results offer new perspectives for future research and lay the foundation for further development of artificial intelligence applications in the field of water consumption prediction. For future studies, it is recommended to improve model performance using larger datasets, develop real-time prediction systems, and test the model in other geographical regions. Additionally, examining the potential effects of climate change scenarios on water consumption could be among future research topics. The integration of advanced artificial intelligence techniques in water resource management highlights the importance of this study and offers a more dynamic and adaptive approach to water consumption estimation. Moreover, this study highlights the importance of integrating advanced artificial intelligence techniques into water resource management, offering a more dynamic and adaptive approach to water consumption forecasting. The integration of such technologies could significantly improve the efficiency of water distribution networks, ensuring sustainable water use even during periods of high demand. Future advancements in AI models, especially in real-time data processing and climate change adaptation, will be essential to tackle the growing challenges of water scarcity worldwide.