Effective weight optimization strategy for precise deep learning forecasting models using EvoLearn approach.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Time series analysis and prediction are crucial research areas.
• Current prediction models' accuracy heavily relies on their learning process.
• Optimizing learning for accuracy and speed is essential for resource efficiency.

Purpose of the Study:

• To introduce EvoLearn, a novel method for improving and optimizing the learning process of neural-based models.
• To enhance prediction accuracy and reduce learning time in time series forecasting.
• To demonstrate the effectiveness of EvoLearn across various neural network architectures.

Main Methods:

• EvoLearn integrates genetic algorithms with back-propagation for training neural network weights.
• The method selects optimal components from multiple models during training.
• Tested on Multilayer Perceptron (MLP), Deep Neural Networks (DNN), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Gated Recurrent Units (GRU).

Main Results:

• EvoLearn was evaluated on air pollution and energy consumption time series datasets.
• Performance comparison showed EvoLearn significantly improves prediction accuracy over conventional back-propagation.
• A one-tailed paired T-test confirmed the statistical significance of the improvement.

Conclusions:

• EvoLearn offers a superior learning methodology for neural-based time series prediction.
• The proposed method enhances prediction accuracy and optimizes resource usage.
• EvoLearn is a promising framework for accurate time series forecasting.