A Contrastive Study of Machine Learning on Energy Firm Value Prediction

Abstract

For decades, a high prediction error rate of firm value assessment has been reported by using traditional financial evaluation methods, therefore develop a suitable assessment tool to improve firm value prediction accuracy is in urgent. This paper provides a comprehensive review and statistical comparison of six machine learning models: K-Nearest Neighbor, Decision Trees, Support Vector Regression, Artificial Neutral Network, AdaBoost, and Random Forest in oil firm and power firm value prediction. Based on nearly 5000 MA items, this paper finds that for both oil and power industries, the prediction error of ANN is the lowest in all the three measurement terms. ANN performs better than the other five ML models by 18% at least for oil industry, and outperforms the others by 19% for power industry. It shows that ANN models can produce both accurate and reasonably understandable prediction results. ANN can be applied to a wide range of MA decisions and value assessment for energy firms.