Enhanced Real-time Electricity Price Prediction with a Novel Feature Selection Technique

This paper addresses an important classification problem in real-time electricity markets: would the price be high or low? This question is of high interest for hedging and risk management purposes. Instead of trying to predict the exact numerical value of the prices, we propose a three-step statistical feature selection technique. This approach extracts the most significant features for the price classification models and maintains a good prediction performance with very little computational cost. The selected features are applied in four popular data-driven prediction models to classify the future price. The performance of each model is tested using real-world electricity market data. This approach sheds light on a broad class of price prediction problems in electricity markets.