Comparative Analysis of Statistical and Feature Based Time Series Forecasting Models on the BEED Dataset

Abstract

Accurate time series forecasting plays a critical role in data driven decision making across engineering, economics, and industrial domains. Traditional statistical models such as ARIMA and SARIMA have long been employed for univariate forecasting, while recent advances in machine learning and feature based representations have introduced models such as ROCKET and MiniRocket. This paper presents a comprehensive comparative study of classical statistical, probabilistic, and feature based forecasting models applied to the BEED dataset. A univariate time series is constructed from multivariate sensor data and evaluated using ARIMA, SARIMA, Prophet, ROCKET, and MiniRocket models under a unified experimental framework. Performance is assessed using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE). Experimental results demonstrate that SARIMA consistently outperforms other models, highlighting the importance of explicitly modelling seasonality, while feature based approaches exhibit limitations when applied to pure forecasting tasks.