Predicting Enterprise Revenue Change Trends Based on the LightGBM Model to Improve Financial Decision–Making Efficiency

Abstract

Accurate revenue trend prediction is important for engineering enterprises operating in cyclical and technology-driven markets, including manufacturers of electromagnetic materials, antenna substrates, and communication-related components. Traditional financial prediction methods often have limited accuracy and weak adaptability to nonlinear business fluctuations. This study proposes a LightGBM-based framework for predicting enterprise revenue change trends and improving financial decision–making efficiency. Panel data from 326 A-share manufacturing listed companies from 2019 to 2024 are used to construct a three-dimensional indicator system covering financial, business, and macroeconomic variables. Feature engineering is performed using threshold truncation, offset adjustment, and logarithmic transformation, while Bayesian optimization is adopted for parameter tuning. The empirical results show that, compared with XGBoost, the proposed model reduces MAE by 30.5%, shortens training time by 40%, and achieves a prediction direction accuracy of 82.6%. When integrated into budget planning and investment–financing decisions, the model reduces the budget deviation rate by 61.0% and shortens decision response time by 64.2%. The study demonstrates the usefulness of machine learning for financial decision support in manufacturing enterprises and provides a quantitative tool for managing uncertainty in engineering production and electromagnetic-related industrial supply chains.