Financial Crisis Analysis and Prediction in Africa using Machine Learning and Statistical Learning Methods

Abstract

Africa has been prone to financial crises that hinder its development. Understanding the causes and predicting the financial crisis for Africa is important for investors and governments. This article aims to analyze those financial ratios that affect the crises and find the best machine learning model to predict possible financial crises in Africa. 11 different machine learning and statistical learning methods are applied to the African systemic crisis data for 13 African countries from 1860 to 2014 to find the most suitable model for financial crisis prediction. The financial ratios being considered are exchange rate, domestic debt, sovereign external debt, GDP, inflation rate, independence, currency crises and inflation crises. After visualizing various machine learning and statistical learning models, this essay concludes that support vector classifier with the “rbf” kernel is the best model after comparing the performance of those models with metrics that have the priority as follows: accuracy, f1 score, recall and precision. Finally, this article uses the support vector classifier to find that systemic crisis is the financial ratio that have the biggest effect on the African banking crises after shuffling the values of each ratio and analyzing the influence on the model's performance.