A Comparative Simulation Study of Classical and Machine Learning Techniques for Forecasting Time Series Data - Simulation Study Summary

Aug 3, 2026

This manuscript presents a simulation comparison between statistical classical methods and machine learning algorithms for forecasting time series data. Models evaluated include ARIMA, K-Nearest Neighbors (KNN), Support Vector Regression (SVR), and Long-Short Term Memory (LSTM). Performance is assessed using Mean Squared Error (MSE), Mean Absolute Error (MAE), Median Absolute Error, and Root Mean Squared Error (RMSE). Results show KNN and LSTM achieve superior accuracy for medium- and long-term forecasts, while machine learning generally surpasses ARIMA for shorter-term predictions.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages10

Reading Time25 min

AccessFree to download

Introduction

Methodology

ARIMA algorithm

K-nearest neighbors algorithm

Support vector regression algorithm

Long-short term memory (LSTM) algorithm

Simulation setup and data preprocessing

Model comparison using accuracy metrics

Conclusion

Paper—A Comparative Simulation Study of Classical and Machine Learning Techniques for Forecasting… A Comparative Simulation Study of Classical and Machine Learning Techniques for Forecasting Time Series Data [https://doi.org/10.3991/ijoe.v19i08.39853](https://doi.org/10.3991/ijoe.v19i08.39853) Mbarek Iaousse1, Youness Jouilil2(*), Mohamed Bouincha3, Driss Mentagui2 1C3S Laboratory, Hassan II University of Casablanca, Casablanca, Morocco 2Department of Mathematics, Faculty of Sciences, Ibn Tofail University of Kenitra, Kenitra, Morocco 3Faculty of Legal, Economic and Social Sciences of Sale, Mohamed V University of Rabat, Rabat, Morocco [y.jouilil@gmail.com](y.jouilil@gmail.com) Abstract—This manuscript presents a simulation comparison of statistical classical methods and machine learning algorithms for time series forecasting notably the ARIMA model, K-Nearest Neighbors (KNN), The Support Vector Regression (SVR), and Long-Short Term Memory (LSTM) . 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 simulations approve that the KNN and LSTM algorithms have better accuracy than the others models’forecasting notably in the medium and long term. Hence, in the medium and long term, ML models are so powerful on big datasets. However, Machine learning architectures outperform ARIMA for shorter-term predictions. Thus, ARIMA is most appropriate in the case of univariate small data sets, where deep learning algorithms are not yet at their best. Keywords—machine learning, time series forecasting, classical approaches, forecasting 1 Introduction In the recent decade, time series forecasting and analysis have become an important field, especially in medicine, economy, and industry [1] . This paper provides an in-depth examination of the most efficient and widely utilized machine learning algorithms for forecasting. In fact, we propose to conduct a general approach to time series generation. The simulation will be performed to investigate the potential of classical and machine learning algorithms to improve the accuracy of forecasting. We design simulations from stationary models where the residual terms follow the standard normal distribution. Paper—A Comparative Simulation Study of Classical and Machine Learning Techniques for Forecasting… Then, we simulate a large time process data set and compute their corresponding features. The rest of the paper will be structured as follows. We start with the methodology and then the time series generation and data preprocessing. In the third section, we will compare our time series algorithms using different accuracy metrics. In the last, we conclude. 2 Methodology This section exposes the methodology and a review of classical and deep sequential architectures for forecasting time series. Specifically, we consider the following algorithms: ARIMA, SVR, KNN and LSTM architectures. 2.1 ARIMA algorithm Auto-Regressive Integrated Moving Average, or ARIMA, models are a class of models that are used to analyze and forecast time series data [2] . The model is defined mathematically as: (1+ φ1B + φ2 B2 +􀀖+ φp Bp )(1− B)d Xt = (1+ θ1B + θ2 B2 +􀀖+ θqBq ) ηt (1) Where Xt the time series data at time t, B is the backshift operator, d is the order of differencing, θi and ϕi are the parameters of the model, and ηt is the error term. 2.2 K-nearest neighbors algorithm The K-nearest Neighbors (KNN) algorithm is a non-parametric method used for classification and regression. The algorithm tries to find the K observations in the training dataset that are closest to the observation to be predicted, and it assigns the most common output value among those K observations to the observation to be predicted. Mathematically, the KNN algorithm can be represented as: Given a new observation, xnew, and a training

FAQs

Which forecasting models are compared in the simulation study?

The study compares ARIMA, KNN, SVR, and LSTM for time series forecasting.

How is model performance evaluated?

Performance is evaluated using MSE, MAE, Median Absolute Error, and RMSE.

Which models perform better for medium- and long-term forecasting?

KNN and LSTM provide better accuracy for medium- and long-term predictions compared with the other models.

Aug 3, 2026

Upload by Lucas Martin