Sensor-Based Machine Learning PZEM-004T for Energy Consumption Prediction and Anomaly Detection in Support of Sustainable Energy Efficiency

Rafi Aditya Pradana (Universitas Bina Nusantara)
Sani Muhamad Isa (Universitas Bina Nusantara)

AMPLITUDO: Journal of Science & Technology Innovation
Vol. 5 No. 2 (2026): August

Increased electricity consumption requires a smart monitoring system that can predict energy usage and detect anomalies early on to improve energy management efficiency. This study aims to integrate machine learning algorithms with PZEM-004T sensor data for energy consumption prediction and time series-based anomaly detection. Sensor data is processed through preprocessing and modeling stages using Long Short-Term Memory (LSTM) and ARIMA for prediction, as well as Isolation Forest and K-Means for anomaly detection. Prediction performance is evaluated using Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). Based on the test results in Tables 1 and 2, the ARIMA model performed better than the LSTM. This is demonstrated by the consistently lower RMSE and MAE values ​​in both test scenarios. A lower RMSE value indicates a relatively lower deviation from the actual data, while a lower MAE value indicates a smaller average prediction error, resulting in a more accurate energy consumption estimate. The MAPE values for each model configuration are reported in Tables 1, 2, and 3 to support these findings with quantitative evidence.