Developing a neuro-flexible mechanism of bankruptcy risk estimation based on conditional parameters

The object of the study is the estimation of the risk of enterprise bankruptcy. The work is aimed at developing a new model for estimating the risk of enterprise bankruptcy. Estimating the risk of bankruptcy is critical to assessing a company’s financial health. It serves as a key indicator that enables management teams to proactively mitigate potential risks and develop strategies to strengthen the company’s financial position over time. It is possible to enhance our prior bankruptcy prediction model by eliminating the Neural Arithmetic Logic Unit (NALU) block and refining the fuzzifier block to assess if the new architecture can effectively simulate approximate arithmetic for discovering complex financial ratios and relationships. The new model uses our bespoke «neuro-flexible» mechanism that incorporates a fuzzifier block as its initial layer, transforming each financial parameter into a fuzzy representation without any NALU blocks down the line. This approach allows the model to process undefined or missing inputs, enhancing its robustness in varied financial scenarios. The fuzzified values are then processed through linear layers with Mish activation, known for superior generalization performance. Key improvements include optimal categorization of raw numbers through embedding vectors and significant acceleration in learning speed. Experiments conducted using PyTorch on an Apple M1 processor demonstrated a substantial average prediction performance of 72 %, indicating the efficacy of the proposed enhancements in bankruptcy estimation. Bankruptcy risk is important for assessing a company’s financial health. It helps management teams reduce risks and strengthen the company’s finances. By predicting bankruptcy risk, companies can take steps to avoid financial problems and stay in business.