How can Big Data Enhance Machine Learning Model Performance in Predicting Future Prices? The Case of the US Consumer Price Index

The consumer price index (CPI) is known as a key macroeconomic indicator for assessing inflation, and accurate predictions of CPI changes over time are significant for a country's economic development. Here, we collected 100+ years (1913–2025) of CPI for All Urban Consumers (CPI-U) data (monthly percentage change in CPI value) from the United States Bureau of Labor Statistics. We employed multiple forecasting models known as SARIMAX, SARIMA rolling forecast, Holt-Winters (HW), Holt-Winters rolling forecast, Facebook prophet (FP), and Facebook prophet rolling forecast to predict historical, current, and future CPI value. Key findings indicated CPI fell by 27.4% (a one-time event in the last 100 years) during the Great Depression of 1929–1933. Also, both original HW and FP models performed best to accurately predict the seasonal variation in CPI numbers during 2014–2025. Finally, the CPI (seasonally adjusted) value would be around 0.467 (all-year high) in January and 0.422 in April 2025, dropping to an all-year low of (-0.005) in November 2025. As per HW forecast, this pattern (a trend in variation of CPI value) of dropping to the lowest CPI in November and reaching peak (highest) in January would be the same for future years (2025–2027). This significant drop in CPI value in November 2025 could have profound implications for economic policies and planning, as it suggests a potential seasonal deflationary trend. Findings reported here would support the Federal government to adjust its monetary policy to stimulate the economy or adjust the inflation and deflation.