Electricity Price Prediction in the Australian National Electricity Market: A Comparative Study of Machine Learning Models Under Market Volatility

The growing integration of variable renewable energy sources has intensified price volatility in the Australian National Electricity Market (NEM), creating operational challenges for market operators, traders, and battery energy storage systems (BESS). Short-term electricity price forecasting is critical for efficient arbitrage, grid stability, and ancillary services such as Frequency Control Ancillary Services (FCAS). This study presents a comparative evaluation of eight forecasting models: ARIMA, SARIMAX, linear regression, Exponential Smoothing, XGBoost, Long Short-Term Memory (LSTM) networks, and Prophet, with and without exogenous inputs, By using real-time price data from New South Wales (NSW). Model performance is assessed across three distinct market regimes: stable, volatile, and rapidly fluctuating conditions. Results show that XGBoost delivers the most accurate forecasts, achieving the lowest MAE (0.040) and highest \( R^2 \) (0.949), while Prophet with demand-based regressors provides notable improvements under volatile regimes. The findings demonstrate that hybrid and context-aware approaches, which combine market features with non-linear models, offer superior accuracy and robustness, supporting real-time decision-making in electricity markets with high renewable energy penetration.