Paper: arXiv 2412.14529

Authors: Arash Peik, Mohammad Ali Zare Chahooki, Amin Milani Fard, Mehdi Agha Sarram

Abstract

Organizing and managing cryptocurrency portfolios and decision-making on transactions is crucial in this market. Optimal selection of assets is one of the main challenges that requires accurate prediction of the price of cryptocurrencies. In this work, we categorize the financial time series into several similar subseries to increase prediction accuracy by learning each subseries category with similar behavior. For each category of the subseries, we create a deep learning model based on the attention mechanism to predict the next step of each subseries. Due to the limited amount of cryptocurrency data for training models, if the number of categories increases, the amount of training data for each model will decrease, and some complex models will not be trained well due to the large number of parameters. To overcome this challenge, we propose to combine the time series data of other cryptocurrencies to increase the amount of data for each category, hence increasing the accuracy of the models corresponding to each category.

Complexity vs Empirical Score

  • Math Complexity: 7.0/10
  • Empirical Rigor: 6.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced deep learning concepts like Temporal Fusion Transformers and Markov chains for categorization, indicating high mathematical density. While it includes quantitative results from backtests on cryptocurrency data, the lack of explicit implementation details (e.g., code, hyperparameters) or extensive statistical validation slightly limits empirical readiness.

Research Flowchart

  flowchart TD
  A["Research Goal: Improve Cryptocurrency Price Forecasting"] --> B["Step 1: Time Series Categorization<br>(Clustering subseries with similar behavior)"]
  B --> C["Step 2: Cross-Cryptocurrency Data Augmentation<br>(Combine data from other cryptos to increase training set size)"]
  C --> D["Step 3: Model Training<br>(TFT Attention-based models for each category)"]
  D --> E["Outcome 1: Overcame small dataset limitations"]
  D --> F["Outcome 2: Achieved improved prediction accuracy"]
  E --> G["Outcome 3: Enhanced Portfolio Management & Decision Making"]
  F --> G

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes categorizing financial time series into similar subseries and training a separate Temporal Fusion Transformer (TFT) model for each category.

Source sentence
In this work we propose to categorize the financial time series into several subseries with similar behavior and train a TFT model for each category of the subseries separately.

Assumes 2

  • The trading simulation assumes zero exchange fees. execution
    Source sentence
    In these transactions, we considered the 0 exchange fee.
  • Financial covariates are excluded from the TFT models. model
    Source sentence
    In this comparison, we did not include financial covariates in any of the TFT models.

Data

Backtest sample: cryptocurrency · crypto

  • Binance 1-minute transaction data · Binance market-prices proprietary
    2 years · cryptocurrency pairs
    Source sentence
    We collected a dataset of Binance transactions in 1-minute time frames for 2 years, which has not been published before at this granularity.
  • Binance 1-minute transaction data (testing) · Binance market-prices proprietary
    18 months since December 2021 · cryptocurrency pairs
    Source sentence
    We store transaction information of Binance exchange as the largest decentralized cryptocurrency exchange cumulatively in 1-minute time intervals for 18 months since December 2021.

Recognised sources: Binance

Costs

Not modeled A trading or execution result is reported without costs.

Source sentence
In these transactions, we considered the 0 exchange fee.

Finds 4

  • The proposed TFT models with time series categorization achieved more than 6% profit over two weeks, outperforming LSTM and non‑categorised transformers. performance
    Source sentence
    our temporal fusion transformer models with time series categorization have created more than 6% more profit in 2 weeks.
  • If the category selector model achieved 100% accuracy, the strategy would turn 100 USDT into 248 USDT (148% profit) in two weeks. performance
    Source sentence
    if the category selector model achieved 100% accuracy, a profit of 148% could be obtained.
  • With a selector accuracy of 79%, the method achieved a 3.16% profit over a two‑week period. performance
    Source sentence
    with a selector accuracy of 79%, we achieved a profit of 3.16% over a two-week period.
  • The selector models achieved accuracies of 0.79 for the Markov chain selector and 0.74 for the LSTM selector. predictive-accuracy
    Source sentence
    the accuracy of TFT model selector was 0.79 for Markov chain selector and 0.74 for LSTM selector.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 25 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 1 paper in the archive

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (38)

  1. Bitcoin: A peer-to-peer electronic cash system (2008)
  2. G it H ub - arashitc2/ B inance-1-minute-candles
  3. G it H ub - arashitc2/ I mproving-cryptocurrency-price-forecasting-using-time-series-categorization-and-tft
  4. Leveraging latent economic concepts and sentiments in the news for market prediction (2021)
  5. Investigating the informativeness of technical indicators and news sentiment in financial market price prediction (2022)
  6. An Adaptive Multimodal Learning Model for Financial Market Price Prediction (2024) — link
  7. Forecasting mid-price movement of Bitcoin futures using machine learning (2023)
  8. Predicting price changes in ethereum (2019)
and 30 more
  1. Predicting Cryptocurrency Price Returns by Using Deep Learning Model of Technical Analysis Indicators (2023)
  2. Cryptocurrency market trend and direction prediction using Machine Learning: A Comprehensive Survey (2023)
  3. Cryptocurrency price forecasting–A comparative analysis of ensemble learning and deep learning methods (2024)
  4. Stock price prediction based on temporal fusion transformer (2021)
  5. Attention is all you need (2017)
  6. Time-series well performance prediction based on Long Short-Term Memory (LSTM) neural network model (2020)
  7. Temporal fusion transformers for interpretable multi-horizon time series forecasting (2021)
  8. Multi-transformer: A new neural network-based architecture for forecasting S&P volatility (2021)
  9. Neural machine translation by jointly learning to align and translate (2014) — link
  10. Stock price prediction using machine learning techniques (2019)
  11. Hidden Markov model implementation for prediction of stock prices with TF-IDF features (2016)
  12. Stock closing price prediction using machine learning (2019)
  13. Predictive Modeling of Stock Indexes Using Machine Learning and Information Theory (2019)
  14. A comparative study of supervised machine learning algorithms for stock market trend prediction (2018)
  15. Design of stock price prediction model with various configuration of input features (2019)
  16. Stock market forecasting using machine learning: Today and tomorrow (2019)
  17. Cryptocurrencies as a financial asset: A systematic analysis (2019)
  18. Factors influencing cryptocurrency prices: Evidence from bitcoin, ethereum, dash, litcoin, and monero (2018)
  19. Factors affecting cryptocurrency prices: Evidence from ethereum (2020)
  20. Algorithms for C omputer A lgebra (1992)
  21. Software engineering—from auxiliary to key technologies (1992)
  22. Conductive P olymers (1981)
  23. Neuromuscular blocking drugs in man (1976)
  24. Isolation and characterization of plasmid deoxyribonucleic acid from Streptomyces fradiae (1978)
  25. Global integrated drought monitoring and prediction system (GIDMaPS) data sets (2014)
  26. Quantum scissors: teleportation of single-mode optical states by means of a nonlocal single photon (2002)
  27. Mixing induced CP asymmetries in inclusive B decays (1997) — link
  28. deep SIP : deep learning of S upernova I a P arameters (2020)
  29. Dark Energy Survey Year 1 Results: Constraints on Extended Cosmological Models from Galaxy Clustering and Weak Lensing (2019) — link
  30. Fractal dimensional analysis in financial time series (2015)

0 of 38 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.