Forecasting the Profit or Loss Status of Companies using Machine Learning

In the realm of risk assessment for credit allocation to enterprises, financial data is of paramount significance to banking institutions. The profitability of a company serves as a critical indicator for these institutions. Profitable entities are typically perceived as possessing a greater capacity for loan repayment, whereas firms incurring losses are classified into higher-risk categories. Consequently, when determining credit limits, banks conduct thorough examinations of the profit or loss status of businesses. However, as decisions rely on historical and contemporary data, forecasts regarding a company’s future performance are inherently speculative.This study employs machine learning algorithms to enhance the accuracy of predictions concerning the profit or loss status of companies after the forthcoming fiscal year, utilizing current year-end financial data. A dataset consisting of 28,541 year-end records and 22 independent variables was used for this research. The inter-correlations among the independent variables were analyzed, leading to the exclusion of those exhibiting high correlation from the dataset. Outliers were identified and removed using the Z-score method, and the data were standardized to improve model performance. Given the limited number of loss-making firms, the Synthetic Minority Over-sampling Technique (SMOTE) was implemented to rectify the imbalance and facilitate superior outcomes. Following data cleaning and resolving imbalances, various machine learning algorithms including Extreme Gradient Boosting (XGBoost), Decision Tree, Random Forest, and K-Nearest Neighbors (KNN) were deployed. The XGBoost algorithm yielded the most favorable results. Thus, this study illustrates that by leveraging contemporary data, firms can attain valuable insights into their future financial conditions through precise predictions of profit or loss at the end of the impending fiscal year.