Leveraging Dynamic ANN for Early Detection of Financial Distress

Microfinance Institutions (MFIs) play a crucial role in Kenya, contributing approximately 7% to the economy according to the Central Bank of Kenya. Despite their success in promoting financial inclusion, MFIs face challenges such as poor management, competition, and unfavorable business conditions, which can lead to financial distress. In Kenya's economic landscape, financial hardship has been a growing concern since 2019 after the onset of Covid 19, leading to the closure of organizations as they are unable to meet their financial obligations and expectations. This study presents a financial distress prediction dynamic model using Artificial Neural Networks (ANN) using financial ratios as input features where each node represents a single financial metric utilizing their power to measure financial health, stability and standardization across periods of organizations. Data was normalized using a min-max scaling approach, setting values between 0 and 1 for better model convergence and split to training and testing set in a ratio 8:2. The perceived dynamic ANN model utilizes the Multilayer Perceptron Networks (MLP), which is modeled to be adaptable, scalable and adjustable with the ability to self-update its architecture and parameters achieving a 94% accuracy, with strong recall and precision of 92% showing predicted financially distressed instances that were actually distressed. Further, the model demonstrated a ROC-AUC score of 0.99, demonstrating its effectiveness in distinguishing between distressed and non-distressed instances. The model's balanced F1 score of 87% further highlights its value as an Early Warning System (EWS) for financial management, helping organizations make informed decisions early enough to avoid financial crises. To further improve the EWS, it is recommended to integrate a dynamic adjustment mechanism within the ANNGARCH framework. This enhancement aims to capture volatility in predictions and refine accuracy for risk assessment, which will be explored in future research.