Addressing challenges in time series forecasting: a comprehensive comparison of machine learning techniques

The explosion of time series (TS) data, driven by advancements in technology, necessitates sophisticated analytical methods. Modern management systems increasingly rely on analyzing this data, highlighting the importance of efficient processing techniques. State-of-the-art machine learning (ML) approaches for TS analysis and forecasting are becoming prevalent. In this paper, we provide an overview of suitable algorithms for TS regression tasks, comparing their performance with each other and with the traditional autoregressive integrated moving average (ARIMA) method across diverse datasets–including synthetic data, long-term recorded data, data containing outliers, and data with missing values. Our focus is on forecasting accuracy, especially for long-term predictions. A key strength of our work is the comprehensive collection of various ML methods tailored for TS data, along with their evaluation across different datasets that include various challenging scenarios. The results show that tree-based ensemble methods outperform other algorithms in most cases. This study aims to assist researchers and practitioners in selecting the most appropriate algorithm based on specific forecasting requirements and data characteristics.