Improving the accuracy of machine learning models in predicting evacuated tube solar collector through Random Forest replacement

• Experimental investigation of all-glass straight through evacuated tube collector. • Outliers identification by clustering algorithm. • The Random Forest model was employed to calculate and replace the outliers. • Comparison of different methods for replacing outliers. • XGBoost and Random Forest exhibited excellent accuracy in performance prediction of tube. Data is a critical factor that determines the upper limit of machine learning models performance. The quality and quantity of data directly influence a machine learning algorithm’s ability to understand the data and learn patterns. During the operation of solar energy devices, uncertainties such as weather and environmental factors often lead abnormal data that deviate from normal conditions. These low-quality data reduce the prediction accuracy of models. In this work, operational data of a novel all-glass straight through evacuated tube collector were collected through experiments. A clustering algorithm was employed to identify outliers, which were then replaced and imputed using three methods: Random Forest, zero and mean values. Prediction test by machine learning algorithms illustrated that the dataset processed with Random Forest computed values had the highest quality. On this basis, several machine learning models were applied to predict the water outlet temperature and thermal efficiency of the evacuated tube collector. It is indicated that the eXtreme Gradient Boosting (XGBoost) and Random Forest methods achieved higher prediction accuracy, significantly outperforming Decision Tree and Multiple Linear Regression (MLR). Replacing outliers in the original dataset with values estimated by Random Forest algorithm increases the model’s generalization ability and offers a new approach to improve the accuracy of solar energy system performance analysis using machine learning techniques.