Improved Financial Stress Prediction using a Network Analysis and Machine Learning Hybrid Model

Abstract

Anticipating financial hardship is a crucial aspect of financial planning, especially given the current state of uncertainty. This study integrates machine learning and network analysis methodologies to present a new way for predicting financial distress. By employing this methodology, two networks of organisations are formed based on the extent to which their primary financial indicators are similar and interconnected. Unlike the second network, the first network showcases similarity in five distinct attributes rather than just one. During the next stage, seven network-centric attributes are retrieved and then added to the dataset as additional variables. Furthermore, cluster firms undergo community detection techniques, and the resulting labels from these algorithms are included in the cluster as categorical variables. By employing this process, a revised dataset will be generated that incorporates both original and network-derived characteristics. Financial strain can be predicted in three scenarios by using five classification algorithms. During model training, only the initial attributes are utilised at the beginning. The combination of network-centric information from correlation and similarity networks in later instances leads to an enhanced accuracy in the prediction of machine learning models. The improvement can be largely attributed to the capabilities of the similarity network. Moreover, the proposed model has a remarkable predictive ability and offers a comprehensive comprehension of the dynamic interactions among financial institutions. The findings offer policymakers valuable information and demonstrate the efficacy of network-based tactics in constructing predictive models for financial distress.

Keywords: Financial Distress Prediction, Financial Analysis, Network-based Analysis, Machine Learning, Classification, Community Detection