Smart meters are critical for accurate energy monitoring and management, yet they can malfunction, leading to missing values in energy datasets. This study evaluates the performance of various statistical and Machine learning methods for imputing missing values in energy data, specifically focusing on production from photovoltaic (PV) systems and electricity consumption across different building types. This study focus on the Missing Completely at Random (MCAR) mechanism and two distinct scenarios are evaluated: (A) missing values occurring between two existing values, and (B) blocks of 3–5 consecutive missing values. The effectiveness of the methods is assessed as the fraction of missing values increases (5% , 10% and 15%). Additionally, we explored the impact of including up to three previous timesteps (lagged variables) as inputs to machine learning-based imputation methods. The results revealed that this inclusion enhances the performance of these methods. Moreover, machine learning-based imputation methods outperformed statistical techniques, with bagging methods achieving the best results overall.