Dimension Reduction in Economic Forecasting: Discussing and Applying Principal Component Analysis

One of my favorite project from one of the most important concepts in data science: Machine Learning.

This project explores how dimensionality affects predictive accuracy in economic forecasting, specifically for housing prices. Motivated by the challenges posed by the "curse of dimensionality," my group and I analyzed how multiple macroeconomic indicators, including interest rates, inflation, unemployment, GDP, and time, influence a model's performance in predicting the Housing Price Index (HPI).

To address redundancy and overfitting caused by high-dimensional data, I applied Principal Component Analysis (PCA), effectively transforming correlated economic indicators into fewer, more meaningful features. Results demonstrated that utilizing only the first two principal components significantly simplified the model while preserving or even enhancing predictive accuracy. In this project, I cleaned time-series data from the Federal Reserve Economic Database (FRED) via API, engineered features from interest rate, inflation, unemployment rate, GDP, and time index. I also implemented min-max normalization and 5-fold cross-validation for model evaluation, then compared linear regression performance with raw features vs. principal components using PCA

The project combined rigorous empirical testing, theoretical analysis, and modern statistical techniques to provide actionable insights for improving forecasting stability and performance in economic models.