Visual Chart Representations for Cryptocurrency Regime Prediction: A Systematic Deep Learning Study

arXiv cs.CV / 5/5/2026

💬 OpinionIdeas & Deep AnalysisModels & Research

共有:

Key Points

The study systematically tests how different visual encodings of cryptocurrency candlestick data affect regime prediction performance using controlled experiments across Bitcoin, Ethereum, and the S&P 500 from 2018–2024.
It compares multiple image encoding approaches (raw candlestick charts, Gramian Angular Fields, and multi-channel GAF) and various chart-component configurations, showing that simpler inputs often work better.
Among four neural network families (CNN, ResNet18, EfficientNet-B0, and Vision Transformer), a small 4-layer CNN trained on raw candlestick charts achieved the best result with an AUC-ROC of 0.892.
Despite the domain mismatch between natural images and financial charts, ImageNet transfer learning still improves performance by roughly 4–16%, and the paper uses Grad-CAM for interpretability.
The findings suggest that reducing complexity in both representation (e.g., price-only charts) and resolution (e.g., 128×128) can outperform more elaborate, pretrained, or transformer-based setups for visual regime classification.

Abstract

Technical traders have long relied on visual analysis of candlestick charts to identify market patterns and predict price movements. While deep learning has achieved remarkable success in image classification, its application to financial chart images remains underexplored. This paper presents a systematic study comparing different visual representations for cryptocurrency regime prediction. We evaluate three image encoding methods (raw candlestick charts, Gramian Angular Fields, and multi-channel GAF), five chart component configurations, four neural network architectures (CNN, ResNet18, EfficientNet-B0, and Vision Transformer), and the impact of ImageNet transfer learning. Through eight controlled experiments on Bitcoin, Ethereum, and S&P 500 data spanning 2018-2024, we identify optimal configurations for visual regime classification. Our results show that a simple 4-layer CNN on raw candlestick charts achieves 0.892 AUC-ROC, outperforming larger pretrained models. Surprisingly, simpler representations (price-only charts, 128x128 resolution) consistently outperform more complex alternatives. We provide interpretability analysis using GradCAM and demonstrate that transfer learning improves performance by 4-16% despite the domain gap between natural images and financial charts.