Research on Financial Crisis Early Warning based on Improved Backpropagation Neural Network

Now a days, Early Warning Systems (EWS) is significant to predict the financial crises by analyzing economic indicators and market behaviors. However, EWS consists challenges such as lack data availability and difficulty in handling unpredictable events such as geopolitical shocks and pandemics. To overcome these drawbacks, Improved Backpropagation Neural Network (BPNN) is proposed for predicting the in financial risk of enterprises. Initially, the input data is collected financial sectors over 60 companies, then preprocessing is performed using label encoding for converting categorical data into numerical data, also Inter Quartile Range (IQR) method is utilized to handle outliers. Further, Least Absolute Shrinkage and Selection Operator (LASSO) technique identifies these financial indicators by reducing the coefficients of less relevant features to zero. After that, the proposed improved BPNN predicts financial crises by learning patterns from historical financial data and adjusting weights by efficiently identifying complex patterns in financial data. The proposed improved BPNN achieved better results in terms of prediction accuracy such as 62% for safe cases, 22% for light risk, and 16% for severe risk comparison with the existing BPNN.