Enhancing the Accuracy of Conductivity Parameters from Real-Time Rainwater Quality Measurements based on Internet of Things Utilizing Machine Learning

Rainwater, a valuable natural resource, can act as a carrier for various pollutants and ions. Conductivity is one of the leading indicators of water quality, reflecting the presence of dissolved substances acquired during its passage through the atmosphere. Currently, real-time measurement of rainwater quality based on the Internet of Things has been carried out at one measuring station, namely on the rooftop of the Gedung Kuliah Umum (GKU) Telkom University, Indonesia. However, real-time conductivity data has a relatively low level of accuracy when compared with daily laboratory test data. Conductivity data from real-time measurements during rain has two different conditions: when the water temperature is uncontrolled and controlled. Therefore, this research aims to increase accuracy and obtain the best general model in various conditions using machine learning. The methods used are Artificial Neural Network (ANN), Support Vector Regression (SVR), and Extreme Gradient Boosting (XGBoots). Next, each model result is evaluated for Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Relative Error (MRE), and Rsquared (R²). Based on the model evaluation results, the XGBoost model performs best in increasing accuracy. XGBoost Grid Search with RMSE results close to zero, a very high R² of 0.99, and a relatively low MAE value of 3.5, which shows very accurate prediction capabilities. They were followed by SVR RBF Random Search, which also shows good performance with low RMSE and R² of 0.23 and 0.78 and MAE value of less than 0.16. Meanwhile, ANN epoch 100 has comprehensive performance but still needs to improve prediction accuracy compared to the other two models. Therefore, XGBoost with Grid Search can be the primary choice for improving accuracy based on the evaluation results given.