Abstract
The precise identification of financial patterns is crucial for predicting market trend and making trading decisions. However, few AI-empowered studies have considered heterogeneous visual patterns in financial data, computer vision, and machine learning in combination to tackle this challenge. To address this limitation, we identify the most common chart patterns in financial time series data and formulate them as a nine-class classification problem. Our study systematically investigates diverse combinations of feature extractors, descriptors, and machine learning classifiers to enhance classification accuracy across these complex financial pattern categories. Our goal is to perform comparison between feature extractors (ORB, SIFT, A-KAZE) and descriptors (BRIEF, FREAK, ORB, SIFT, A-KAZE) and machine learning classifiers (LightGBM, SVM, KNN, Random Forest and Decision Tree). The achieved results show that the combination of ORB, SIFT and LightGBM can get the best accuracy, with 90.6%. These results might have important implications in the realization of AI-empowered trading algorithms and for the accuracy of an algorithmic trading system.
Keywords
Subject Areas
Related YTU Activities
Sustainability activities and projects connected to this article.
