Risk Assessment for the Electricity Retail Market Based on Decision Tree Algorithm

Abstract

Driven by deregulation and the widespread adoption of renewable energy, the electricity retail market exhibits significant coupling among price volatility risks, load uncertainty risks, and market settlement risks faced by power retailers. Traditional risk assessment methods based on parametric assumptions and linear approximations struggle to capture nonlinear interactions between risk factors, while insufficient model interpretability limits decision-makers’ trust in and adoption of early warning outcomes. This study proposes a risk assessment framework for electricity retail markets employing an improved CART decision tree algorithm. By constructing a three-dimensional risk indicator system encompassing wholesale price fluctuations, user-side load characteristics, and power retailer transaction behaviors, the model utilizes the Gini coefficient minimization criterion to achieve recursive segmentation of risk features and optimization of node purity. The analysis identifies three primary risk drivers – day-ahead-to-real-time price differentials, peak-to-valley load ratios, and medium-to-long-term contract proportions – with a cumulative contribution rate of 46.9%. In binary risk prediction tasks, the model achieved an accuracy of 87.8% ± 2.1%, recall of 84.5% ± 2.8%, and F1 score of 86.1% ± 2.3% on the test set, surpassing logistic regression and support vector machines by 9.8 and 7.2 percentage points respectively. Through decision path tracing and rule extraction, the model generates explicit risk assessment rules providing actionable mitigation strategies for power retailers. The study demonstrates that decision tree algorithms offer significant advantages in balancing predictive accuracy and model transparency, effectively supporting dynamic risk management in electricity retail markets.