Quantile Regression Machine Learning Techniques to Handle Outliers in Time Series Data

Abstract

Time series data often exhibit outliers that can significantly distort the results of traditional regression methods, leading to inaccurate forecasts and suboptimal decision-making. Quantile regression offers a robust alternative by estimating the conditional median or other quantiles of the response variable, thus providing a more comprehensive analysis of the underlying data distribution. This review paper explores the integration of quantile regression with advanced machine learning techniques to effectively handle outliers in time series data. By comparing different approaches, we evaluate how well they perform in terms of root mean square error (RMSE) and mean absolute error (MAE). According to the comparison graph, the EMD-QRnet method greatly boosted the MAE and RMSE by 84.3% and 1.00%, which is the highest gain of any method when compared to others.