Paper: arXiv 2509.10542

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

Abstract

Precise short-term price prediction in the highly volatile cryptocurrency market is critical for informed trading strategies. Although Temporal Fusion Transformers (TFTs) have shown potential, their direct use often struggles in the face of the market’s non-stationary nature and extreme volatility. This paper introduces an adaptive TFT modeling approach leveraging dynamic subseries lengths and pattern-based categorization to enhance short-term forecasting. We propose a novel segmentation method where subseries end at relative maxima, identified when the price increase from the preceding minimum surpasses a threshold, thus capturing significant upward movements, which act as key markers for the end of a growth phase, while potentially filtering the noise. Crucially, the fixed-length pattern ending each subseries determines the category assigned to the subsequent variable-length subseries, grouping typical market responses that follow similar preceding conditions. A distinct TFT model trained for each category is specialized in predicting the evolution of these subsequent subseries based on their initial steps after the preceding peak. Experimental results on ETH-USDT 10-minute data over a two-month test period demonstrate that our adaptive approach significantly outperforms baseline fixed-length TFT and LSTM models in prediction accuracy and simulated trading profitability. Our combination of adaptive segmentation and pattern-conditioned forecasting enables more robust and responsive cryptocurrency price prediction.

Complexity vs Empirical Score

  • Math Complexity: 7.0/10
  • Empirical Rigor: 5.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced deep learning architectures (Temporal Fusion Transformers) with specialized algorithms (adaptive segmentation and pattern-conditioned categorization), indicating significant mathematical complexity. It also provides empirical validation on real cryptocurrency data with trading simulations, demonstrating solid implementation and data-heavy experimentation.

Research Flowchart

  flowchart TD
  A["Research Goal: Enhance Short-Term Cryptocurrency Price Prediction in Volatile Markets"] --> B["Data Input: ETH-USDT 10-Minute Price Data"]
  B --> C["Adaptive Segmentation Strategy<br>Split data at relative maxima<br>Capture key growth phases"]
  C --> D["Pattern-Based Categorization<br>Group subseries by preceding fixed-length patterns"]
  D --> E["Category-Specific TFT Modeling<br>Train distinct TFT models per category"]
  E --> F["Computational Outcome: Multi-Model Forecasting System"]
  F --> G["Key Findings: Superior Accuracy & Trading Profitability<br>Outperforms Fixed-Length TFT & LSTM Baselines"]

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 an Adaptive Temporal Fusion Transformer modeling approach that uses dynamic, threshold‑based subseries segmentation and pattern‑conditioned forecasting with specialized TFT models for each pattern category.

Source sentence
an adaptive TFT modeling approach leveraging dynamic subseries lengths and pattern-based categorization to enhance short-term forecasting.

Assumes 3

  • Zero transaction fees are assumed in the simulated trading strategy. execution
    Source sentence
    We assume zero transaction fees for simplicity.
  • A subseries ends only when the price rise from the last minimum exceeds a predefined threshold Th. model
    Source sentence
    ... only if the rise from the last identified minimum vtmin,last to this peak vtpeak exceeds a predefined threshold Th.
  • The closing price, transformed into a volatility rate series, is sufficient as the core feature for modeling. data
    Source sentence
    Similar to our previous work [14], the core feature driving our analysis and modeling is derived from the closing price.

Data

  • ETH‑USDT 10‑minute Binance dataset · Binance market-prices public
    December 27, 2021 to November 22, 2024 · ETH‑USDT pair
    Source sentence
    The empirical basis for this investigation is a proprietary, high‑resolution dataset of 1‑minute ETH‑USDT trading intervals, collected from the Binance exchange for the period of December 27, 2021, to November 22, 2024.

Recognised sources: Binance

Costs

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

Source sentence
We assume zero transaction fees for simplicity.

Finds 3

  • The Adaptive TFT model achieves 51.36% accuracy and 51.11% precision, outperforming all baseline models on the ETH‑USDT test set. predictive-accuracy
    Source sentence
    As shown in Table 1, the proposed Adaptive TFT model achieves an accuracy of 51.36% and a precision of 51.11%, outperforming all baseline models.
  • In simulated trading, the Adaptive TFT strategy ends with a final asset value of 117.22 USDT, higher than Buy‑and‑Hold (108.32 USDT) and other baseline strategies. performance
    Source sentence
    Our Adaptive TFT model resulted in a final asset value of 117.22 USDT, demonstrating higher profitability compared to both the passive Buy and Hold strategy and the trading strategies based on the baseline predictive models.
  • The Adaptive model shows a recall of 92.31% but a low specificity of 19.28%, reflecting an aggressive upward‑trend prediction bias. statistical
    Source sentence
    Our adaptive model exhibits an exceptionally high recall (92.31%), indicating the successful identification of nearly all true upward price movements. This bullish conviction, however, comes at the cost of a low specificity (19.28%).

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 46 related papers). Coverage is partial, so this is neither replication nor refutation.

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 (57)

  1. Leveraging Time Series Categorization and Temporal Fusion Transformers to Improve Cryptocurrency Price Forecasting — → 2412.14529
  2. Timexer: Empowering transformers for time series forecasting with exogenous variables (2024)
  3. Enhancing Cryptocurrency Price Prediction by Applying Temporal Fusion Transformers on Subseries (2025)
  4. k-shape: Efficient and accurate clustering of time series (2015)
  5. Regime-switching factor investing with hidden Markov models (2020)
  6. Bitcoin: A peer-to-peer electronic cash system (2008)
  7. G it H ub - arashitc2/ A daptive-TFT
  8. G it H ub - arashitc2/ B inance-1-minute-candles
and 49 more
  1. G it H ub - arashitc2/ I mproving-cryptocurrency-price-forecasting-using-time-series-categorization-and-tft
  2. Leveraging latent economic concepts and sentiments in the news for market prediction (2021)
  3. Investigating the informativeness of technical indicators and news sentiment in financial market price prediction (2022)
  4. An Adaptive Multimodal Learning Model for Financial Market Price Prediction (2024) — link
  5. Financial Market Prediction using Contrastive Learning -based News Representation and Multi - Resolution Market Data (2025)
  6. Forecasting mid-price movement of Bitcoin futures using machine learning (2023)
  7. Predicting price changes in ethereum (2019)
  8. Predicting Cryptocurrency Price Returns by Using Deep Learning Model of Technical Analysis Indicators (2023)
  9. Cryptocurrency market trend and direction prediction using Machine Learning: A Comprehensive Survey (2023)
  10. Cryptocurrency price forecasting–A comparative analysis of ensemble learning and deep learning methods (2024)
  11. Stock price prediction based on temporal fusion transformer (2021)
  12. Attention is all you need (2017)
  13. Time-series well performance prediction based on Long Short-Term Memory ( LSTM ) neural network model (2020)
  14. Rhine: A regime-switching model with nonlinear representation for discovering and forecasting regimes in financial markets (2024)
  15. A deep fusion model for stock market prediction with news headlines and time series data (2024)
  16. MFB: A Generalized Multimodal Fusion Approach for Bitcoin Price Prediction Using Time-Lagged Sentiment and Indicator Features (2025)
  17. Temporal fusion transformers for interpretable multi-horizon time series forecasting (2021)
  18. Multi-transformer: A new neural network-based architecture for forecasting S&P volatility (2021)
  19. Neural machine translation by jointly learning to align and translate (2014) — link
  20. Stock price prediction using machine learning techniques (2019)
  21. Hidden Markov model implementation for prediction of stock prices with TF-IDF features (2016)
  22. Stock closing price prediction using machine learning (2019)
  23. Predictive Modeling of Stock Indexes Using Machine Learning and Information Theory (2019)
  24. A comparative study of supervised machine learning algorithms for stock market trend prediction (2018)
  25. Design of stock price prediction model with various configuration of input features (2019)
  26. Stock market forecasting using machine learning: Today and tomorrow (2019)
  27. Cryptocurrencies as a financial asset: A systematic analysis (2019)
  28. Factors influencing cryptocurrency prices: Evidence from bitcoin, ethereum, dash, litcoin, and monero (2018)
  29. Factors affecting cryptocurrency prices: Evidence from ethereum (2020)
  30. Algorithms for C omputer A lgebra (1992)
  31. Software engineering—from auxiliary to key technologies (1992)
  32. Conductive P olymers (1981)
  33. Neuromuscular blocking drugs in man (1976)
  34. Isolation and characterization of plasmid deoxyribonucleic acid from Streptomyces fradiae (1978)
  35. Global integrated drought monitoring and prediction system (GIDMaPS) data sets (2014)
  36. Quantum scissors: teleportation of single-mode optical states by means of a nonlocal single photon (2002)
  37. Mixing induced CP asymmetries in inclusive B decays (1997) — link
  38. deep SIP : deep learning of S upernova I a P arameters (2020)
  39. Cryptocurrency price prediction with convolutional neural network and stacked gated recurrent unit (2022)
  40. Informer: Beyond efficient transformer for long sequence time-series forecasting (2021)
  41. A new approach to the economic analysis of nonstationary time series and the business cycle (1989)
  42. A Markov model of switching-regime ARCH (1994)
  43. Markov-switching GARCH models in R: The MSGARCH package (2019)
  44. Time-varying transition probabilities for Markov regime switching models (2017)
  45. Darts: User-Friendly Modern Machine Learning for Time Series (2022) — link
  46. Dynamic modeling and forecasting of time-evolving data streams (2019)
  47. Regime shifts in streams: Real-time forecasting of co-evolving time sequences (2016)
  48. Dark Energy Survey Year 1 Results: Constraints on Extended Cosmological Models from Galaxy Clustering and Weak Lensing (2019) — link
  49. Fractal dimensional analysis in financial time series (2015)

1 of 57 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.