Comparing the Accuracy of Classical and Machine Learning Methods in Time Series Forecasting: A Case Study of USA Inflation

Abstract

This paper presents a comparison of statistical classical methods and machine learning algorithms for time series forecasting notably the Exponential Smoothing, hybrid ARIMA-GARCH model, K-Nearest Neighbors (KNN), Prophet, and Long-Short Term Memory (LSTM). The data set used in this study is related to US inflation and covers the period from 1965 to 2021. The performance of the models was evaluated using different metrics especially Mean Squared Error (MSE), Mean Absolute Error (MAE), Median Absolute Error (Median AE), and Root Mean Squared Error (RMSE). The results of the numerical comparison show that the best performance was achieved by Exponential Smoothing, followed closely by KNN. The results indicate that these two models are well-suited for forecasting inflation in the US. ARIMA-GARCH, LSTM, and Prophet performed relatively poorly in comparison. Overall, the findings of this study can be useful for practitioners in choosing the most suitable method for forecasting inflation in the US in the short-term period.