Deep Belief Kronecker Forward Harmonic Net for Financial Distress Prediction using Federated Learning

Main Article Content

Eega Varnika
G. Madhavi

Abstract

Due to the globalization of the world financial system, the prediction of financial distress is essential for enterprises, especially for financial organizations. Accurate bankruptcy prediction is needed for every financial organization to make careful financing decisions. Owing to financial regulations, the prediction of financial distress becomes complex. For resolving the complexity, the Federated Learning (FL) approach is utilized. FL improves the data privacy and eliminates the security threats, and it is required for the institutions to control the risks in large-scale applications. Hence, this paper devises the FL-based Deep Belief Kronecker Forward Harmonic Net (FL_DBKFHNet) for Financial Distress Prediction (FDP). The servers and nodes are considered as the major part of the FL approach. The local training is carried out using the local data from each device, and updates are passed to the server. After completing the aggregation in the server, the upgraded data is downloaded by the global model. For each epoch, the updated training is iterated in the local model via the utilization of downloaded data. The FDP is done in the training model, where the input data is utilized to detect the financial distress using the proposed FL_DBKFHNet. Moreover, FL_DBKFHNet-based FDP achieved better accuracy, loss function, Mean Square Error (MSE), Root MSE, and Mean Average Precision of 93.99%, 0.251, 0.066, 0.258, and 0.915.

Article Details

Section
Articles