Predicting stock prices on the Tehran Stock Exchange by a new hybridization of Fuzzy Inference System and Fuzzy Imperialist Competitive Algorithm

چکیده مقاله

This paper presents a hybrid model for stock price prediction on the Tehran Stock Exchange by combining a Fuzzy Inference System (FIS) with a Fuzzy Imperialist Competitive Algorithm (FICA). The proposed approach, named FICA+FIS, utilizes two fuzzy systems: one to dynamically tune the parameters of the FICA based on search strategy factors, and another to predict stock prices using ten key economic variables, including lowest and highest stock prices, trading volume, market indices, dollar exchange rate, gold price, and oil price. The model employs triangular membership functions with linguistic variables—Low, Medium, and High—for both input and output, and the fuzzy rules for the inference engine are optimized by the FICA, which itself is parameterized through a separate fuzzy tuner.

The study evaluates the model using datasets from six active companies in the Tehran Stock Exchange from 2010 to 2015, applying 10-fold cross-validation for training and testing. The performance of FICA+FIS is compared against two groups of methods: first, hybrid models using classic Imperialist Competitive Algorithm, Genetic Algorithm, and Whale Optimization Algorithm with FIS; and second, traditional machine learning techniques including neural networks, support vector machines, decision trees, random forests, and Gaussian process regression. Results demonstrate that the fuzzy-enhanced FICA outperforms its classical counterpart and other metaheuristic hybrids in obtaining fuzzy rule sets. Moreover, the FICA+FIS model shows superior predictive accuracy and stability, with the lowest deviation between training and testing errors across most datasets.

A key achievement of this research is the effective integration of fuzzy logic with an evolutionary metaheuristic to handle the nonlinearity and uncertainty inherent in stock market data. The FICA+FIS model not only achieves high prediction quality but also offers better interpretability through its fuzzy rule base, unlike black-box models such as neural networks. Additionally, the dynamic parameter tuning of FICA via a fuzzy system addresses a common challenge in metaheuristics—fixed parameter settings—leading to improved exploration and exploitation balance. Despite longer computational times compared to some classical methods, the model's robustness and low generalization error make it a reliable tool for financial forecasting.

In conclusion, the proposed hybrid FICA+FIS framework provides a novel and effective methodology for stock price prediction in volatile markets. It successfully combines the adaptability of fuzzy systems with the optimization power of an enhanced imperialist competitive algorithm, yielding interpretable and accurate forecasts. Future work may focus on feature selection, multi-objective optimization, and application to other financial markets to further enhance the model's efficiency and scope.