複數模糊類神經於趨向類別預測之研究

面對現今的國際化環境，投資已成為許多企業與人們的獲利方式，股票為此領域相當普遍的交易模式，但股價波動所牽涉層面廣泛，固然難以估計與預測，人工智慧中的深度學習即為當今預測的最佳工具之一。本研究提出一種新形態之複數模糊類神經分類模型（Complex Neuro-Fuzzy Classification Model; CNFC），以減法分群演算法（Subtractive Clustering Algorithm; SCA）識別資料趨向類別，輔助模型進行動態式分類預測，其中採用粒子群最佳化演算法（Particle Swarm Optimization; PSO）與遞迴最小平方法（Recursive Least Squares Estimator; RLSE）為複合式最佳化演算法（Hybrid optimization algorithm），針對模型不同部分的參數進行優化，將有效提升模型優化效率。透過複數高斯模糊集合的特性，模糊化輸入資料的類別隸屬程度，更加精確描述類別值域，增強模型的預測及應用能力。實驗透過重複性與文獻多樣化模型比較結果，驗證CNFC的預測效能與PSO-RLSE的最佳化成效於股價時間序列資料具有較佳能力。

關鍵字

複數模糊類神經；複數模糊集；特徵選取；粒子群最佳化演算法；遞迴最小平方法

並列摘要

Purpose-Facing the current international environment, investment has become a way of profit for many businesses and people, stocks are a common method of trading in this area, but stock price fluctuations have a wide range of influences, it is difficult to estimate and forecast. Deep learning in Artificial Intelligence (AI) is one of the best tools for current prediction. Design/methodology/approach-This study proposes a novel Complex Neuro- Fuzzy Classification Model (CNFC), identifying the directional classification of data by Subtractive Clustering Algorithm (SCA) and assisting models for dynamic classification prediction. The model uses Particle Swarm Optimization (PSO) and Recursive Least Squares Estimator (RLSE) as the hybrid optimization algorithm for parameters optimization of different parts of the model will effectively improve the efficiency of model optimization. Findings-The experiments to verify the predictive performance of CNFC and the optimization effect of PSO-RLSE through repetitive and literature diversification models, which has better ability in stock price time series data. Research limitations/implications-In this study, the class degree of input data is fuzzified through the characteristics of complex Gaussian fuzzy sets, which more accurately describes the class value and enhances the prediction and application ability of the model. In the future, we plan to combine the famous classifier (e.g., SVM, Softmax) with the CNFC model. Practical implications - In this study, we provide an innovative stock price forecasting model that can be used as an auxiliary investment tool for investors, and financial practitioners can further explore the relationship between stock prices and the overall economy. Originality/value-This study is the first attempt to optimize the parameters of the CNFC model using PSO-RLSE hybrid optimization algorithm.