Accurate daily energy production prediction of solar Photovoltaic (PV) systems is essential for improving efficiency and integration of the PV systems with the energy grid. This work focuses on addressing the technical challenges associated with the prediction of daily energy production using data from a solar PV system by integrating variable importance with the k-nearest neighbors (k-nn) algorithm. Specifically, we propose a combined methodology by using the Pearson correlation coefficient to determine weighted feature-based distances of important features as part of the k-nn method. The proposed methodology improves on the conventional k-nn method by allocating adaptive weights to input features/variables, such as timestamp and meteorological parameters, based on their correlation with solar PV energy production. The performance of the proposed methodology is examined against Naïve forecasting model and conventional equal-weight k-nn model for a 264 kWp PV system. The findings obtained highlight the importance of considering the weighted impact of operational conditions on energy production prediction, thus advancing sustainable energy management practices.