Artificial neural networks applied to forecasting time series.

Area of Science:

• Computer Science
• Machine Learning
• Data Science

Background:

• Artificial Neural Networks (ANN) are increasingly utilized for complex forecasting tasks.
• Time series forecasting is crucial across various scientific and industrial domains.
• Evaluating different ANN architectures is essential for optimizing predictive accuracy.

Purpose of the Study:

• To describe and compare key Artificial Neural Network (ANN) models for time series forecasting.
• To establish a standard procedure for applying ANNs in time series analysis.
• To evaluate the practical performance of Multilayer Perceptron (MLP), Radial Base Function (RBF), Generalized Regression Neural Network (GRNN), and Recurrent Neural Network (RNN) models.

Main Methods:

• Analysis of four primary ANN models: MLP, RBF, GRNN, and RNN.
• Application of these models to a time series dataset comprising 244 data points.
• Comparative performance evaluation based on forecasting error metrics.

Main Results:

• All four analyzed ANN models achieved a forecasting error of less than 10%.
• The Radial Base Function (RBF) network exhibited the superior performance.
• Recurrent Neural Network (RNN) and Multilayer Perceptron (MLP) models showed strong, comparable results, outperforming the Generalized Regression Neural Network (GRNN).

Conclusions:

• ANN models demonstrate a close fit and robust forecasting capacity for time series data.
• The RBF network is recommended as the top-performing model for this specific time series forecasting task.
• Further research should explore ANN advantages, limitations, and potential solutions for enhanced future applications.