MACHINE LEARNING–ENABLED EARLY WARNING SYSTEM FOR DETECTING MICRO-INFLATION CLUSTERS IN THE U.S. ECONOMY

Abstract

Inflation in the United States has become noticeably more erratic in recent years, and this has pushed researchers, policymakers, and anyone watching the economy to look for ways to spot early shifts in pricing pressure before they show up in the main CPI numbers. This study builds a machine learning early warning system that looks for the small pockets of rising prices that often form beneath the surface. It uses high-frequency CPI and PPI data, volatility indicators, forecast residuals from several time-series models, and a set of anomaly detection tools to catch signs that inflation is starting to take shape. The system draws on familiar forecasting models like ARIMA, exponential smoothing, and Prophet to produce forward-looking residual features that flag unexpected movements in price patterns. These residuals are blended with unsupervised anomaly scores, rolling momentum signals, and supervised models to estimate the chance that localized inflation pressure is developing months in advance. The modeling pipeline is backed by a range of robustness checks that look at how sensitive the system is to different CPI definitions, alternative PPI group choices, shifting lead times, and the structural changes that took place around the COVID period. SHAP values are used to break down the predictions of the supervised models, showing which features drive the alerts most strongly. The results highlight the influence of residual anomalies, persistent PPI momentum, and volatility spikes, which consistently show up as warning signals. Backtests indicate that combining residual features with momentum indicators improves performance relative to simple rules that many analysts still rely on, especially when predicting CPI accelerations three months ahead. The improvement is most visible after 2020, a period when producer price shocks move more directly into consumer prices and are easier for the system to detect. The results here show that mixing forecast residuals with engineered PPI and CPI features offers a practical way to spot early signs of inflation. The system remains understandable, holds up across several forms of stress testing, and can highlight brewing inflationary pressure before it is visible in the headline indexes. This provides a solid base for building early warning tools that help with economic monitoring, policy analysis, and risk management during times when price behavior becomes uncertain and harder to track.