Akademik Veri Yönetim Sistemi
EXPERT SYSTEMS WITH APPLICATIONS, cilt.296, 2026 (SCI-Expanded, Scopus)
Financial time series forecasting is a crucial but challenging undertaking that is necessary for creating successful strategies. Reliable forecasting systems are vital for effective investment management and algorithmic trading. This necessity drives extensive research into methods for machine learning and statistics. In the literature, numerous studies have focused on predicting price direction in stock time series, with most aiming to forecast the price direction on the same day or the following day. However, there are relatively few studies that approach this prediction by segmenting the time series into smaller parts, and the application of fuzzy sets in this context remains largely unexplored. This study presents a new method for predicting the next day's price direction, distinguishing it from previous research approaches. The approach involves identifying local trends in time series data, extracting features from these trends, converting these features into fuzzy values, and then prediction by using a deep learning method. The 40 stocks listed on BIST (Borsa Istanbul) and their daily closing values are included in the study over a ten-year period. The proposed method predicted the next day price direction as the average of 40 stocks with 79.04% accuracy score.