Comprehensive analysis of electricity price dynamics and influencing factors in deregulated markets based on multi-year data for strategic planning and risk management

Electricity prices in deregulated markets are influenced by multiple interdependent factors, including demand fluctuations, fuel costs, renewable energy penetration, carbon pricing, and weather variability, making their prediction highly challenging. The volatile and non-stationary nature of these prices often leads to overfitting, poor generalization, and convergence to local minima in predictive models. This study develops and evaluates three machine learning models—K-Nearest Neighbors Regression (KNNR), Extra Trees Regression (ETR), and CatBoost Regression (CATR)—along with their hybrid variants optimized using three nature-inspired algorithms: the Starfish Optimization Algorithm (SOA), the Dollmaker Optimization Algorithm (DOA), and the Ladybug Beetle Optimization Algorithm (LBOA). Hybridization leverages mechanisms such as adaptive population control, mutation diversification, and roulette-wheel selection to enhance hyperparameter tuning and improve exploration of the search space. All models are rigorously tested using 5-fold cross-validation and evaluated with R 2 , RMSE, MARE, MBE, and U95 metrics to assess accuracy, stability, and convergence. Results demonstrate that the LBOA-enhanced self-attention CatBoost (CASA) consistently surpasses baseline and other hybrid models, achieving R 2 above 0.97 and RMSE below 2.5 across training, validation, and test datasets. The paper demonstrates how well adaptive metaheuristic optimization and attention mechanisms work together to provide a reliable, scalable, and comprehensible framework for simulating intricate power market dynamics.