Pengembangan Sensor Pengukuran pH Virtual untuk Sistem Hidroponik Biaya Rendah Berbasis IoT dengan KNN dan

Abstract

Intelligent hydroponics systems, incorporating IoT and machine learning, can become a precision agriculture solution. However, physical water pH sensors are expensive, and a low-cost solution can be proposed using virtual sensors. Our research propose a pH virtual sensor for Hydroponics IoT using secondary data from Kaggle. We employ decision tree, Extreme Gradient Boosting (XGBoost), and K-Nearest Neighbors (KNN) regression algorithms for prediction using data from the TDS meter and temperature sensor as input features. Some metrics for testing are Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R² score. Moving average method to enhance the model performance and validate the data using Pearson correlation coefficient (PCC) matrix. Test results show that moving average with a window size of 181 reduces the RMSE of the KNN model from 0.98 to 0.30. KNN achieves a more optimal R² score (0.91) than the decision tree (0.90) and XGBoost (0.57). Virtual pH sensors with KNN are the most valid, as the correlation values of pH with TDS and water temperature are-0.34 to-0.35 and- 0.44 to-0.31, compared to the secondary dataset. This virtual sensor can be a low-cost solution in IoT hydroponics because can reduce product costs from USD 400 to USD 10. Keywords: virtual sensors, pH sensor, low-cost, Internet of Things, hydroponic