Comparing A Hybridization of Fuzzy Inference System and Particle Swarm Optimization Algorithm with Deep Learning to Predict Stock Prices

چکیده مقاله

This research paper presents a comparative study of two intelligent models for stock price prediction in the Tehran Stock Exchange: a hybrid fuzzy inference system optimized by particle swarm optimization (PSO) and a deep learning model. The study aims to capture the complex relationships between ten key economic variables—such as lowest price, highest price, trading volume, USD exchange rate, global gold and oil prices, and market indices—and the stock prices of nine active companies. The hybrid fuzzy-PSO model combines a Mamdani-type fuzzy inference system with PSO to optimize the fuzzy rule base, while the deep learning model employs a feed-forward neural network with five hidden layers and 37 neurons, trained using the Adam optimizer.

The results indicate that the deep learning model outperforms both the hybrid fuzzy-PSO model and a traditional artificial neural network in terms of prediction accuracy. Specifically, the deep learning model achieved approximately 79% better performance than the fuzzy-PSO model and 92% better than the standard neural network across the tested companies. This superior accuracy highlights the capability of deep learning architectures to model complex, non-linear patterns in financial time-series data, making it a powerful tool for stock market forecasting.

Despite its lower predictive accuracy, the hybrid fuzzy-PSO model offers several notable advantages. It demonstrates more stable and consistent behavior across different companies, with significantly lower variance in its predictions compared to the deep learning and neural network models. Additionally, the fuzzy-PSO model converges faster during training, indicating greater computational efficiency in rule optimization. Importantly, the fuzzy system provides interpretable rules that can be analyzed and understood by market analysts—a valuable feature for practical decision-making in financial markets.

The study also validates the models on two distinct datasets: the first includes three companies from earlier years, and the second covers six companies from a more recent period. In both cases, the deep learning model maintained its accuracy advantage, while the fuzzy-PSO model showed robustness and reliability, particularly for certain companies like Iran Copper, Persian Gulf Petrochemical, and Lotus. The research underscores the trade-off between accuracy and interpretability in financial predictive modeling and suggests that the choice of model may depend on the specific needs of the analyst—whether for high precision or for understandable, rule-based insights.

Overall, this paper makes a meaningful contribution to the field of financial forecasting by implementing and comparing advanced hybrid and deep learning approaches in an emerging market context. It successfully demonstrates the potential of deep learning for high-accuracy predictions while affirming the value of fuzzy logic systems for stability, speed, and interpretability in stock market analysis.