<![CDATA[Solar energy is one of the most sustainable renewable energy sources, and its utilization through photovoltaic (PV) systems is highly dependent on the accuracy of solar irradiance prediction. However, predicting solar irradiance remains challenging due to the nonlinear relationship between weather variables and solar cell data, as well as the high dimensionality of input features. Therefore, this study aims to develop an efficient prediction model using a hybrid Convolutional Neural Network–Long Short-Term Memory (CNN–LSTM) architecture combined with SelectKBest feature selection. Two heterogeneous datasets, consisting of solar cell data and weather data, were integrated and processed through preprocessing, Z-score normalization, and feature engineering, resulting in 28 input features. Subsequently, F-regression-based SelectKBest was applied to select 13 of the most relevant features. The performance of the CNN–LSTM model with and without feature selection was evaluated using Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), coefficient of determination (R²), and training time. The results show that feature selection reduced training time from 151.14 seconds to 138.32 seconds and improved MAPE from 45.10% to 42.86%. However, RMSE increased from 69.46 to 73.27 and R² slightly decreased from 0.937 to 0.930. These findings indicate that SelectKBest can improve computational efficiency while maintaining acceptable predictive performance in solar irradiance forecasting.]]>
