Classifier Comparison for Recession Prediction Using Imbalanced Financial Data

Recession predictions are essential for policymakers, businesses, and investors to prepare for economic downturns. However, economic datasets are often highly imbalanced with recessions being a less frequent event, making it challenging for machine learning (ML) models to make accurate predictions. This research explored various ways to use machine learning (ML) classifiers for recognizing financial recessions and forecasting them one quarter ahead, addressing the particular challenge of data imbalance. This study began with a raw dataset containing quarterly economic indicators, which were then extended into a monthly time-series dataset through interpolation methods to match the monthly recession data. Various machine learning algorithms such as Random Forest (RF), Logistic Regression (LR), Decision Tree (DT), Support Vector Machines (SVM), and Multilayer Perceptron (MLP) were applied to recognition and forecasting financial recession using Monte Carlo simulations. This study systematically evaluated the performance of various classifiers by applying different data balancing techniques, including classic bootstrapping, weighted bootstrapping, and SMOTE (Synthetic Minority Oversampling TEchnique) to both original and interpolated datasets. Additionally, principal component analysis (PCA) was applied to explore its impact on classifier performance. This study not only evaluated the impact of different data balancing techniques but also analyzed various classifiers' performance before and after applying PCA, providing a comprehensive evaluation of balancing methods and dimensionality reduction, offering valuable insights for economic forecasting and data-driven decision-making.