Accurate current modelingCurrent modeling is critical for energy management, fault detection, and system optimization. This study evaluates the performance of long short-term memory (LSTM) networks and autoregressive integrated moving average (ARIMA) models for predicting current in systems with square-tooth waveforms. Current data, collected every 15 min over six months from a smart meter in Shah Alam, Selangor, Malaysia, exhibited periodic sharp transitions between high and low states. LSTM outperformed ARIMA in capturing these abrupt transitions and periodicity due to its ability to learn temporal dependencies and handle nonlinear patterns. ARIMA, designed for linear trends and stationary data, struggled to model the sharp transitions and showed limited accuracy in replicating the waveform’s structure. The comparative analysis highlights LSTM’s advantages in accuracy and robustness, making it well-suited for complex time-series data such as electrical currents. While ARIMA remains useful for simpler datasets, its limitations in nonlinearity reduce its applicability in this context. The findings underscore the importance of selecting appropriate modeling techniques based on data characteristics and demonstrate the potential of LSTM in advanced energy management systems. This study provides a foundation for leveraging machine learning models for accurate and efficient predictive analytics in electrical systems.