Hybrid ANFIS-MPA and FFNN-MPA Models for Bitcoin Price Forecasting.

Area of Science:

• Computational intelligence
• Financial forecasting
• Time series analysis

Background:

• Accurate short-term Bitcoin price prediction remains challenging.
• Existing forecasting models often struggle with the volatility of cryptocurrency markets.
• Optimization algorithms are crucial for training complex predictive models.

Purpose of the Study:

• To develop and evaluate hybrid forecasting models for short-term Bitcoin price prediction.
• To integrate the Marine Predators Algorithm (MPA) with Adaptive Neuro-Fuzzy Inference Systems (ANFIS) and Feed-Forward Neural Networks (FFNN).
• To compare the performance of MPA-enhanced models against other metaheuristic algorithms.

Main Methods:

• Daily Bitcoin price data from 2022 was utilized.
• Data was transformed into supervised time-series structures with various input configurations.
• Two hybrid models, ANFIS-MPA and FFNN-MPA, were developed and tested.
• Performance was evaluated against six established metaheuristic training algorithms.

Main Results:

• MPA demonstrated superior performance, achieving lower prediction errors and faster convergence.
• The hybrid ANFIS-MPA and FFNN-MPA models consistently outperformed baseline algorithms.
• Models showed reliable performance across different complexities and robust, reproducible results with low variance.

Conclusions:

• The Marine Predators Algorithm is an effective optimizer for neuro-fuzzy and neural network models in financial time-series forecasting.
• Hybrid models integrating MPA offer enhanced predictive accuracy and stability for Bitcoin price prediction.
• The proposed ANFIS-MPA and FFNN-MPA approaches represent a significant advancement in financial forecasting techniques.