Pith. sign in

REVIEW 4 major objections 6 minor 71 references

crypto price prediction using lstm+xgboost

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims a two-stage LSTM+XGBoost model beats five standalone baselines on Bitcoin, Ethereum, Litecoin, and Dogecoin price forecasting.

desk verdict A standard LSTM+XGBoost hybrid whose only experimental evidence is a single table with no protocol; the claimed gains are small, the architecture is prior art, and the paper should be desk-rejected. read the letter →

arxiv 2506.22055 v1 pith:2PGLTHTS submitted 2025-06-27 cs.LG

Mehul Gautam This is my paper · ORCID
classification cs.LG
keywords cryptocurrencypricepredictionLSTMXGBoosthybridmodeltimeseriesforecastingMAPEMinMaxRMSE
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that cryptocurrency price forecasting improves when a Long Short-Term Memory network and an XGBoost model are chained in two stages rather than used alone. The LSTM is asked to learn the temporal dependencies in a rolling window of price data, and XGBoost is then asked to turn the LSTM's final hidden state into a price prediction. On Bitcoin, Ethereum, Litecoin, and Dogecoin, the author reports that the hybrid achieves the lowest test MAPE (0.0488) and lowest Min-Max normalized RMSE (0.0659) among six models, ahead of LSTM, CNN, Transformer, ARIMA, and XGBoost. The practical interest is that a modest architectural pairing could give traders and analysts a more accurate tool in a notoriously volatile market.

What carries the argument

The carrying mechanism is a two-stage pipeline in which an LSTM, a recurrent neural network with gated memory, first compresses a rolling input window $X=[x_{(t-n+1)}, \ldots, x_{(t)}]\in\mathbb{R}^{n\times d}$ into a single learned vector. The cell's forget, input, output, and candidate gates update a memory state $C_t$ and hidden state $h_t$, and only the final hidden state $z=h_n\in\mathbb{R}^k$ (with $k=64$) is passed onward. Stage two is an XGBoost regressor that learns a mapping $\hat{y}=f(z)=\sum_{m=1}^M f_m(z)$ by minimizing the regularized objective $\sum_i \ell(\hat{y}_i,y_i)+\sum_m \Omega(f_m)$, with $\Omega(f)=\gamma T + \frac{1}{2}\lambda\sum_j w_j^2$ penalizing tree leaves and leaf weights. The point of the chain is that XGBoost never sees raw prices; it sees a learned temporal summary, so the two models are doing complementary work.

What would settle it

Run all six models on an identical, fixed train/validation/test split of the same price histories, with the same hyperparameters and multiple random seeds; if the hybrid's MAPE of 0.0488 does not beat XGBoost alone's 0.0533 by more than run-to-run variation, the reported advantage does not hold.

Watch

Extended reading notes

Core claim

The paper's central claim is that the hybrid LSTM+XGBoost architecture outperforms every standalone model it is compared with on the task of predicting next-period cryptocurrency prices. In the reported comparison, the hybrid's test MAPE of 0.0488 and MinMax RMSE of 0.0659 are the lowest of the six models, beating the closest competitor, XGBoost alone, which posts 0.0533 and 0.0705 respectively. The author presents this as evidence that the two-stage design combines LSTM's strength at modeling sequential structure with XGBoost's strength at handling nonlinear feature interactions. The claim is posed as generalizable across the four major cryptocurrencies tested and across global and localized exchange data, although the paper's own limitation section notes that overfitting, limited explainability, and dependence on historical price features remain open concerns.

Load-bearing premise

The whole comparison stands on the assumption that all six models were evaluated on the same unseen test data with no leakage from the test period into training or model selection, a protocol the paper does not actually specify.

Editorial extensions

If this is right

  • If the reported comparison is taken at face value, anyone building a crypto price forecaster from price history alone would do better to chain LSTM with XGBoost than to tune either model alone: the hybrid's MAPE is 0.0488 versus 0.0533 for the best standalone model.
  • The ordering of baselines (Transformer at 0.0594 MAPE ahead of CNN at 0.0612 but behind XGBoost at 0.0533) suggests that on this dataset, tree-based nonlinear regression on tabular features is a stronger baseline than deep sequence models.
  • Because the metrics are scale-independent (MAPE and MinMax RMSE), the claimed improvement is intended to transfer across the four coins' very different absolute price levels.
  • The two-stage design keeps the output interpretable at the XGBoost stage, so feature-importance tools can be applied to the latent representation produced by the LSTM.
  • The paper's own forward-looking section expects the same hybrid pattern to work with Transformer encoders and with external sentiment features, which is a direct extension of the claimed architecture.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • An extension the paper leaves implicit: because the reported margins are small (0.0045 MAPE over XGBoost alone), a reader should treat 'outperforms' as a single-run point estimate unless repeated trials show the gap is stable.
  • The abstract promises sentiment and macroeconomic features, but the dataset description lists only price, volume, and market cap; feeding exogenous variables through the same two-stage pipeline is a natural test the paper does not run.
  • The normalized metrics make the same pipeline applicable to other high-volatility assets; testing the model on altcoins outside the top four or on commodities would show whether the claimed adaptability is a property of the architecture or of these four coins.
  • The paper notes potential overfitting in its limitation section; a concrete hedge would be to evaluate on a walk-forward split that re-trains the model each month, which would also clarify whether the hybrid advantage survives regime changes.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper proposes a two-stage hybrid model in which an LSTM extracts temporal features from cryptocurrency price series and XGBoost performs regression on the LSTM's final hidden state. It gives the standard LSTM and XGBoost equations, reports exploratory data analysis for Bitcoin, Ethereum, Dogecoin, and Litecoin, and presents Table II comparing six models on MAPE and MinMax RMSE. The conclusion states that the LSTM+XGBoost hybrid outperforms standalone models across the four cryptocurrencies and claims the model integrates global and localized exchange data with sentiment and macroeconomic auxiliary features.

Significance. If the empirical claim were properly supported, the contribution would be modest and incremental: the LSTM+XGBoost two-stage architecture is already described in prior cited work, and the reported performance edge over the next-best model is small. The paper's positive aspects are its explicit metric definitions and the internally consistent ordering in Table II. However, the absence of any reproducible experimental protocol and the direct contradiction between the claimed auxiliary features and the stated limitations mean the manuscript does not currently establish its central claim or its claimed novelty.

major comments (4)
  1. [§IV-A, §IV-J, Table II] The manuscript never states the data source, date range, sampling frequency, train/validation/test split, or preprocessing for the four cryptocurrencies, nor does it specify hyperparameters (beyond k=64), n_steps_in, n_steps_out, XGBoost M/γ/λ, number of runs, or random seeds. Table II reports single-point test metrics; the hybrid's edge over XGBoost is 0.0045 in MAPE and 0.0046 in MinMax RMSE, which is well within plausible run-to-run variability for LSTM training. Consequently the central claim that the hybrid 'outperforms standalone models' is not supported as written.
  2. [§I, §V-A.3] The abstract and Section I state that XGBoost models 'auxiliary features such as sentiment scores and macroeconomic indicators' and that the model is evaluated on 'global and localized exchange data,' but Section V-A.3 explicitly says sentiment and geopolitical events 'are not accounted for in the current model,' and Section IV-A lists only OHLCV and marketcap features. The model actually evaluated is not the model described, and the claimed sentiment, macroeconomic, and localized-exchange contributions are absent from the experiments.
  3. [Table II, §V] The conclusion claims the model 'consistently outperforms' across Bitcoin, Ethereum, Litecoin, and Dogecoin, yet Table II reports only aggregate test metrics with no per-currency breakdown, confidence intervals, or statistical significance tests. The table is therefore insufficient to support a per-currency or 'consistent' outperformance claim.
  4. [§IV-J, Eq. (14)] The MinMax RMSE definition uses max(A)-min(A) over the actual values, but the paper does not state whether this range is computed on the training portion only or on the full dataset, nor does it describe the temporal split. A random split of financial time series can leak future information into training, and a full-data range mixes test information into the evaluation normalization; both choices need to be specified and justified.
minor comments (6)
  1. [§III, Eqs. (1)-(12)] The quantities n_steps_in, n_steps_out, d, and the XGBoost regularization parameters in Eqs. (9)-(11) are introduced but never given values; please provide a full hyperparameter table.
  2. [§V] The manuscript contains two sections numbered V ('Conclusion and Novelty' and 'Limitations and Future Research Work'); the second should be renumbered, for example as Section VI.
  3. [References] References [23]-[28] duplicate [16]-[21], and the in-text citation [22] in Section III does not match the cited survey; the bibliography should be cleaned up and deduplicated.
  4. [§I] The template placeholders 'Index Terms—IEEE, template, style, LaTeX' and 'Identify applicable funding agency here. If none, delete this.' remain in the text and should be removed.
  5. [§V, Key Novel Contributions] The 'Enhanced Interpretability' bullet lists a 'Potential integration' as a key contribution; either implement and report it or move it to future work.
  6. [Figures] Figure captions are present for Figures 1-12 but the figures themselves are not visible in the provided text; please confirm that the final PDF includes all figures and that each is legible.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the empirical comparison in Table II is not defined in terms of its own conclusion.

full rationale

The paper's central claim is an empirical performance comparison (Table II: Hybrid LSTM+XGBoost MAPE 0.0488 vs. XGBoost 0.0533). This claim is not produced by a circular derivation chain: the LSTM cell equations (Eqs. 2-7) and XGBoost objective (Eqs. 9-11) are standard textbook definitions, and the evaluation metrics (Eqs. 13-14) are conventional definitions that do not encode the hybrid model's superiority. No parameter of the model is fitted to the reported test metric and then renamed as a prediction; the test numbers are presented as results of training and evaluation rather than as identities. The paper contains no load-bearing self-citations: reference [22] supplies the standard LSTM gate equations, and the hybrid-architecture citations [7], [21] are prior external works, not the present author's. There is a serious reproducibility problem - the manuscript omits dataset source, date range, train/validation/test split, hyperparameters, and run-to-run variance, so the 0.0045 MAPE margin cannot be distinguished from noise - and an internal inconsistency between the abstract's claim of sentiment/macroeconomic features and Section V-A.3's admission that 'cryptocurrency prices are also influenced by non-quantitative signals such as social media trends, news sentiment, and geopolitical events, which are not accounted for in the current model.' Both of these are correctness/evidence concerns, not circularity: nothing in the paper's equations or citations reduces the outperformance claim to its own assumption. Accordingly, no circular step is identified and the circularity score is 0.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The paper contributes no new mathematical objects. It relies on standard LSTM and XGBoost formulations plus an unstated evaluation-protocol assumption. The many unspecified hyperparameters and split choices are the main hidden degrees of freedom. The abstract's promised sentiment, macroeconomic, and localized data are effectively phantom inputs because they never appear in the model description or experiments.

free parameters (5)
  • LSTM hidden state size k = 64
    Eq. (8) sets k=64 'as per model design', with no tuning or sensitivity analysis reported.
  • n_steps_in (look-back window length)
    Defined in Eq. (1) but never given a value; it directly controls what the LSTM sees.
  • n_steps_out (forecast horizon)
    Mentioned after Eq. (12) as the output horizon, but never specified.
  • XGBoost tree count and regularization (M, gamma, lambda)
    Equations (10)-(11) define these, but no values are reported for the experiments behind Table II.
  • Train/test split ratio and method
    No split is described. The entire comparison depends on this unstated choice.
assumptions (4)
  • standard math LSTM update equations in Eqs. (2)-(7) are the standard recurrent network formulas.
    These are textbook LSTM equations from Hochreiter and Schmidhuber; accepted as background.
  • standard math XGBoost minimizes the regularized loss in Eq. (10) over an ensemble of trees.
    This is the standard XGBoost objective from Chen and Guestrin; accepted as background.
  • domain assumption Historical OHLCV and market capitalization data contain predictive signal for future prices.
    The whole forecasting exercise presupposes that past price and volume patterns generalize to future prices; the paper offers no evidence for this.
  • ad hoc to paper The evaluation protocol is identical and leakage-free across all compared models.
    Nowhere stated in the paper. Without this assumption, Table II is meaningless, and the paper gives no way to check it.

how reviews work

0 comments
Cite this review

Pith. "Pith review of crypto price prediction using lstm+xgboost." pith.science (2026). https://pith.science/paper/2PGLTHTS

@misc{pith2026250622055,
  author       = {Pith},
  title        = {Pith review of: crypto price prediction using lstm+xgboost},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2PGLTHTS}},
  note         = {Machine review of arXiv:2506.22055}
}
read the original abstract

The volatility and complex dynamics of cryptocurrency markets present unique challenges for accurate price forecasting. This research proposes a hybrid deep learning and machine learning model that integrates Long Short-Term Memory (LSTM) networks and Extreme Gradient Boosting (XGBoost) for cryptocurrency price prediction. The LSTM component captures temporal dependencies in historical price data, while XGBoost enhances prediction by modeling nonlinear relationships with auxiliary features such as sentiment scores and macroeconomic indicators. The model is evaluated on historical datasets of Bitcoin, Ethereum, Dogecoin, and Litecoin, incorporating both global and localized exchange data. Comparative analysis using Mean Absolute Percentage Error (MAPE) and Min-Max Normalized Root Mean Square Error (MinMax RMSE) demonstrates that the LSTM+XGBoost hybrid consistently outperforms standalone models and traditional forecasting methods. This study underscores the potential of hybrid architectures in financial forecasting and provides insights into model adaptability across different cryptocurrencies and market contexts.

Figures

Figures reproduced from arXiv: 2506.22055 by the authors.

Figure 3
Figure 3. 30-Day Rolling Volatility of Cryptocurrencies [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 1
Figure 1. Cryptocurrency Closing Prices Over Time To facilitate a comparative analysis, we also normalized the prices to a common starting point, as shown in [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 4
Figure 4. Daily Returns Distribution for Each Cryptocurrency [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figures from the paper (6 more)
Figure 2
Figure 2. Figure 2: Normalized Cryptocurrency Prices (First day = 100) [PITH_FULL_IMAGE:figures/full_fig_p004_2.png]
Figure 5
Figure 5. Figure 5: Trading Volume Over Time The relationship between price and volume is depicted in [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 8
Figure 8. Figure 8: illustrates each cryptocurrency’s market share evolution. Bitcoin has remained dominant in terms of market cap, though its share fluctuates. This supports the findings of Vidal￾Tomas et al. [35]. ´ F. Correlation Analysis Correlation between cryptocurrencies is vital f…
Figure 9
Figure 9. Figure 9: Cryptocurrency Price Correlation Matrix [PITH_FULL_IMAGE:figures/full_fig_p005_9.png]
Figure 11
Figure 11. Figure 11: Time Series Decomposition for Bitcoin Price [PITH_FULL_IMAGE:figures/full_fig_p006_11.png]
Figure 12
Figure 12. Figure 12: SMA Crossover Strategy vs Buy and Hold for Bitcoin [PITH_FULL_IMAGE:figures/full_fig_p006_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

71 extracted references · 65 canonical work pages

  1. [7]

    Hybrid deep learning and machine learning model for cryptocurrency prediction,

    Wang, Y ., & Zhang, L., “Hybrid deep learning and machine learning model for cryptocurrency prediction,”Expert Systems with Applications, vol. 183, 115437, 2021

    work page 2021

  2. [22]

    A survey of forecasting meth- ods based on deep learning and their applications in financial time series,

    Hu, Z., Zhao, Y ., & Khushi, M., “A survey of forecasting meth- ods based on deep learning and their applications in financial time series,”Expert Systems with Applications, vol. 183, 115410, 2021. https://doi.org/10.1016/j.eswa.2021.115410

    work page arXiv 2021

  3. [51]

    Attention-based CNN-LSTM and XGBoost hybrid model for stock prediction

    Z. Shi, Y . Hu, G. Mo, and J. Wu, “Attention-based CNN-LSTM and XGBoost hybrid model for stock prediction,”arXiv preprint arXiv:2204.02623, 2022

    work page Pith review arXiv 2022

  4. [52]

    Optimizing financial time series predictions with hybrid ARIMA, LSTM, and XGBoost models,

    R. Nichani, L. Gasmi, N. Laiche, and S. Kabou, “Optimizing financial time series predictions with hybrid ARIMA, LSTM, and XGBoost models,”Studies in Engineering and Exact Sciences, vol. 5, no. 2, 2025. [Online]. Available: https://ojs.studiespublicacoes.com.br/ojs/index.php/ sees/article/view/11188

    work page 2025

  5. [23]

    Long short-term memory,

    Hochreiter, S., & Schmidhuber, J., “Long short-term memory,” Neural Computation, vol. 9, no. 8, pp. 1735–1780, 1997. https://doi.org/10.1162/neco.1997.9.8.1735

    doi:10.1162/neco.1997.9.8.1735 1997

  6. [24]

    Brownlee, J.,Deep Learning for Time Series Forecasting, Machine Learning Mastery, 2018

    work page 2018

  7. [1]

    Bitcoin: A Peer-to-Peer Electronic Cash System,

    Nakamoto, S., “Bitcoin: A Peer-to-Peer Electronic Cash System,” 2008

    work page 2008

  8. [2]

    What are the main drivers of the Bitcoin price? Evidence from wavelet coherence analysis,

    Kristoufek, L., “What are the main drivers of the Bitcoin price? Evidence from wavelet coherence analysis,”PLoS ONE, vol. 10, no. 4, e0123923, 2015

    work page 2015

Show all 71 references
  1. [3]

    Deep learning with long short-term memory networks for financial market predictions,

    Fischer, T., & Krauss, C., “Deep learning with long short-term memory networks for financial market predictions,”European Journal of Oper- ational Research, vol. 270, no. 2, pp. 654–669, 2018

    2018

  2. [4]

    A CNN–LSTM model for gold price time-series forecasting,

    Livieris, I. E., Pintelas, E., & Pintelas, P., “A CNN–LSTM model for gold price time-series forecasting,”Neural Computing and Applications, vol. 32, no. 23, pp. 17351–17360, 2020

    2020

  3. [5]

    XGBoost: A scalable tree boosting system,

    Chen, T., & Guestrin, C., “XGBoost: A scalable tree boosting system,” KDD ’16: Proceedings of the 22nd ACM SIGKDD International Con- ference on Knowledge Discovery and Data Mining, pp. 785–794, 2016

    2016

  4. [6]

    Stock price prediction using hybrid models,

    Singh, A., & Srivastava, S., “Stock price prediction using hybrid models,”Procedia Computer Science, vol. 167, pp. 2281–2290, 2020

    2020

  5. [8]

    Predicting the price of Bitcoin using machine learning,

    McNally, S., Roche, J., & Caton, S., “Predicting the price of Bitcoin using machine learning,” PDP 2018, pp. 339–343, 2018

    2018

  6. [9]

    Cryptocurrency price prediction using tweet volumes and sentiment analysis,

    Abraham, J., Higdon, D., Nelson, J., & Ibarra, J., “Cryptocurrency price prediction using tweet volumes and sentiment analysis,”SMU Data Science Review, vol. 1, no. 3, pp. 1–22, 2018

    2018

  7. [10]

    Cryptocurrency Price Prediction Using ML Techniques: A Comparative Study,

    Al-Yahya, M., & Alsudairi, M., “Cryptocurrency Price Prediction Using ML Techniques: A Comparative Study,”Journal of King Saud University - Computer and Information Sciences, 2022

    2022

  8. [11]

    Box, G.E.P., Jenkins, G.M., & Reinsel, G.C.,Time Series Analysis: Forecasting and Control, Wiley, 2015

    2015

  9. [12]

    Time series forecasting using a hybrid ARIMA and neural network model,

    Zhang, G.P., “Time series forecasting using a hybrid ARIMA and neural network model,”Neurocomputing, vol. 50, pp. 159–175, 2003

    2003

  10. [13]

    Predicting stock market index using fusion of machine learning techniques,

    Patel, J., Shah, S., Thakkar, P., & Kotecha, K., “Predicting stock market index using fusion of machine learning techniques,”Expert Systems with Applications, vol. 42, no. 4, pp. 2162–2172, 2015

    2015

  11. [14]

    A survey of the applications of text mining in financial domain,

    Kumar, A., & Ravi, V ., “A survey of the applications of text mining in financial domain,”Knowledge-Based Systems, vol. 114, pp. 128–147, 2016

    2016

  12. [15]

    Temporal Fusion Transformers for interpretable multi-horizon time series forecasting,

    Lim, B., Arik, S. ¨O., Loeff, N., & Pfister, T., “Temporal Fusion Transformers for interpretable multi-horizon time series forecasting,” International Journal of Forecasting, vol. 37, no. 4, pp. 1748–1764, 2021

    2021

  13. [25]

    Stock price prediction via discov- ering multi-frequency trading patterns,

    Zhang, Y ., Aggarwal, C., & Qi, G., “Stock price prediction via discov- ering multi-frequency trading patterns,”Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp. 2141–2149, 2017. https://doi.org/10.1145/3097983.3098008

    2017

  14. [26]

    A hybrid machine learning model for stock market forecasting,

    Ghoshal, S., & Roberts, S., “A hybrid machine learning model for stock market forecasting,”Expert Systems with Applications, vol. 157, 113476, 2020

    2020

  15. [27]

    A hybrid deep learning model com- bining LSTM and XGBoost for stock price forecasting,

    Wang, W., Zhang, L., & Li, M., “A hybrid deep learning model com- bining LSTM and XGBoost for stock price forecasting,”Mathematics, vol. 10, no. 3, 413, 2022. https://doi.org/10.3390/math10030413

    2022 doi

  16. [28]

    Forecasting the volatility of cryptocurrency time series using LSTM and XGBoost,

    Kim, Y ., & Won, C. H., “Forecasting the volatility of cryptocurrency time series using LSTM and XGBoost,”Applied Sciences, vol. 8, no. 11, 2086, 2018. https://doi.org/10.3390/app8112086

    2018 doi

  17. [29]

    Forecasting and trading cryptocur- rencies with machine learning under changing market conditions,

    Sebasti ˜ao, H., and Godinho, P., “Forecasting and trading cryptocur- rencies with machine learning under changing market conditions,” Financial Innovation, vol. 7, no. 1, pp. 1–30, 2021. https://doi.org/10. 1186/s40854-020-00217-x

    2021

  18. [30]

    A comparative study of Bitcoin price prediction using deep learning,

    Ji, S., Kim, J., and Im, H., “A comparative study of Bitcoin price prediction using deep learning,”Mathematics, vol. 7, no. 10, p. 898,

  19. [31]

    Conditional tail-risk in cryptocurrency markets,

    Borri, N., “Conditional tail-risk in cryptocurrency markets,”Journal of Empirical Finance, vol. 50, pp. 1–19, 2019. https://doi.org/10.1016/j. jempfin.2018.11.002

    2019 doi

  20. [32]

    V olatility spillover effects in leading cryptocurrencies: A BEKK-MGARCH analysis,

    Katsiampa, P., Corbet, S., and Lucey, B., “V olatility spillover effects in leading cryptocurrencies: A BEKK-MGARCH analysis,”Finance Research Letters, vol. 29, pp. 68–74, 2019. https://doi.org/10.1016/j.frl. 2019.03.009

    2019 doi

  21. [33]

    Can volume predict Bitcoin returns and volatility? A quantiles-based approach,

    Balcilar, M., Bouri, E., Gupta, R., and Roubaud, D., “Can volume predict Bitcoin returns and volatility? A quantiles-based approach,”Economic Modelling, vol. 64, pp. 74–81, 2017. https://doi.org/10.1016/j.econmod. 2017.03.019

    2017 doi

  22. [34]

    The economics of BitCoin price formation,

    Ciaian, P., Rajcaniova, M., and Kancs, d’A., “The economics of BitCoin price formation,”Applied Economics, vol. 48, no. 19, pp. 1799–1815,

  23. [35]

    Herding in the cryp- tocurrency market: CSSD and CSAD approaches,

    Vidal-Tom ´as, D., Ib´a˜nez, A. M., and Farin´os, J. E., “Herding in the cryp- tocurrency market: CSSD and CSAD approaches,”Finance Research Letters, vol. 30, pp. 181–186, 2019. https://doi.org/10.1016/j.frl.2018. 09.008

    2019 doi

  24. [36]

    Co-explosivity in the cryptocurrency market,

    Bouri, E., Shahzad, S. J. H., and Roubaud, D., “Co-explosivity in the cryptocurrency market,”Finance Research Letters, vol. 29, pp. 178–183,

  25. [37]

    V olatility connectedness in the cryptocurrency market: Is Bitcoin a dominant cryptocurrency?,

    Yi, S., Xu, Z., and Wang, G. J., “V olatility connectedness in the cryptocurrency market: Is Bitcoin a dominant cryptocurrency?,”In- ternational Review of Financial Analysis, vol. 60, pp. 98–114, 2018. https://doi.org/10.1016/j.irfa.2018.08.012

    2018 doi

  26. [38]

    Estimating the volatility of cryptocurrencies during bearish markets by employing GARCH models,

    Kyriazis, N. A., Daskalou, K., Arampatzis, M., Prassa, P., and Papaioan- nou, E., “Estimating the volatility of cryptocurrencies during bearish markets by employing GARCH models,”Heliyon, vol. 5, no. 8, e02239,

  27. [39]

    https://doi.org/10.1016/j.frl.2018.07.005

    2018 doi

  28. [40]

    A comparison of ARIMA and LSTM in forecasting time series,

    Siami-Namini, S., Tavakoli, N., and Namin, A. S., “A comparison of ARIMA and LSTM in forecasting time series,” in17th IEEE Interna- tional Conference on Machine Learning and Applications (ICMLA), pp. 1394–1401, 2018. https://doi.org/10.1109/ICMLA.2018.00227

    2018

  29. [41]

    An ensemble of LSTM deep learning networks for forecasting cryptocurrency pricing using technical indicators,

    Kumar, D., Rath, S. K., Mahapatra, R. P., and Kumar, M., “An ensemble of LSTM deep learning networks for forecasting cryptocurrency pricing using technical indicators,”IEEE Access, vol. 9, pp. 106291–106308,

  30. [42]

    https://doi.org/10.1016/j.heliyon.2019.e02239

    2019 doi

  31. [43]

    Technical trading and cryptocurrencies,

    Hudson, R., and Urquhart, A., “Technical trading and cryptocurrencies,” Annals of Operations Research, vol. 297, pp. 191–220, 2021. https://doi. org/10.1007/s10479-019-03357-1

    2021 doi

  32. [44]

    Cryptocurrency price drivers: Wavelet coherence analysis revisited,

    Phillips, R. C., and Gorse, D., “Cryptocurrency price drivers: Wavelet coherence analysis revisited,”PLoS ONE, vol. 13, no. 4, e0195200,

  33. [45]

    Asymmetric and time-frequency spillovers among commodities, cryptocurrencies, and conventional assets,

    Naeem, M. A., Karim, S., Jamasb, T., and Nepal, R., “Asymmetric and time-frequency spillovers among commodities, cryptocurrencies, and conventional assets,”Energy Economics, vol. 107, 105512, 2022. https://doi.org/10.1016/j.eneco.2021.105512

    2022

  34. [46]

    Cryptocurrency forecasting with deep learning chaotic neural networks,

    Lahmiri, S., and Bekiros, S., “Cryptocurrency forecasting with deep learning chaotic neural networks,”Chaos, Solitons & Fractals, vol. 118, pp. 35–40, 2019. https://doi.org/10.1016/j.chaos.2018.11.014

    2019 doi

  35. [47]

    An ad- vanced CNN-LSTM model for cryptocurrency forecasting,

    Livieris, I. E., Kiriakidou, N., Stavroyiannis, S., and Pintelas, P., “An ad- vanced CNN-LSTM model for cryptocurrency forecasting,”Electronics, vol. 10, no. 3, p. 287, 2021. https://doi.org/10.3390/electronics10030287

    2021 doi

  36. [48]

    Co-movements between Bitcoin and Gold: A wavelet coherence analysis,

    Kang, S. H., McIver, R. P., and Hernandez, J. A., “Co-movements between Bitcoin and Gold: A wavelet coherence analysis,”Physica A: Statistical Mechanics and its Applications, vol. 536, 120888, 2019. https://doi.org/10.1016/j.physa.2019.04.124

    2019 doi

  37. [49]

    A hybrid deep learning model for time series forecasting based on LSTM and Gated Recurrent Unit,

    Deng, S., He, Z., Xu, D., “A hybrid deep learning model for time series forecasting based on LSTM and Gated Recurrent Unit,”Expert Systems with Applications, vol. 194, 116540, 2022. https://doi.org/10. 1016/j.eswa.2022.116540

    2022

  38. [50]

    Combining feature selection and deep learning for cryptocurrency price prediction,

    Zhang, X., Wang, Z., and Liu, D., “Combining feature selection and deep learning for cryptocurrency price prediction,”Neural Computing and Applications, vol. 34, pp. 1035–1050, 2022. https://doi.org/10.1007/ s00521-021-06358-7

    2022

  39. [53]

    A novel cryptocurrency price trend forecasting model based on LightGBM,

    Sun, X., Liu, M., and Sima, Z., “A novel cryptocurrency price trend forecasting model based on LightGBM,”Finance Research Letters, vol. 32, 101084, 2020. https://doi.org/10.1016/j.frl.2018.12.032

    2020 doi

  40. [54]

    Predicting the direction, maximum, minimum and closing prices of daily Bitcoin exchange rate using machine learning techniques,

    Mallqui, D. C. A., and Fernandes, R. A. S., “Predicting the direction, maximum, minimum and closing prices of daily Bitcoin exchange rate using machine learning techniques,”Applied Soft Computing, vol. 75, pp. 596–606, 2019. https://doi.org/10.1016/j.asoc.2018.11.038

    2019 doi

  41. [55]

    Price Prediction of the Cryptocurrency from Niche Market Based on Random Forest, LSTM and XGBoost,

    S. Tian, “Price Prediction of the Cryptocurrency from Niche Market Based on Random Forest, LSTM and XGBoost,”Advances in Eco- nomics, Management and Political Sciences, vol. 87, pp. 97–106, 2024

    2024

  42. [56]

    Stock Price Prediction Using a Hybrid LSTM-GNN Model: Integrating Time-Series and Graph-Based Analysis,

    M. S. Sonani, A. Badii, and A. Moin, “Stock Price Prediction Using a Hybrid LSTM-GNN Model: Integrating Time-Series and Graph-Based Analysis,”arXiv preprint arXiv:2502.15813, 2025

    2025 arXiv

  43. [57]

    LSTM-Based Time Series Prediction Model: A Case Study with YFinance Stock Data,

    Y . Yu, “LSTM-Based Time Series Prediction Model: A Case Study with YFinance Stock Data,”ITM Web of Conferences, vol. 70, p. 03015, 2025

    2025

  44. [58]

    Comparing Machine Learning Meth- ods—SVR, XGBoost, LSTM, and MLP—For Forecasting the Moroccan Stock Market,

    H. Oukhouya and K. El Himdi, “Comparing Machine Learning Meth- ods—SVR, XGBoost, LSTM, and MLP—For Forecasting the Moroccan Stock Market,”Computer Sciences & Mathematics Forum, vol. 7, no. 1, p. 39, 2023

    2023

  45. [59]

    Hybrid cryptocurrency price prediction integrating EGARCH and LSTM with explainable AI,

    A. N. M. Mohammed, “Hybrid cryptocurrency price prediction integrating EGARCH and LSTM with explainable AI,”International Journal of Innovative Science and Research Technology, vol. 9, no. 12, 2024. [Online]. Available: https://www.ijisrt.com/ hybrid-cryptocurrency-price-predi...

    2024

  46. [60]

    A Novel Hybrid Approach Using an Attention-Based Transformer + GRU Model for Predicting Cryptocurrency Prices,

    E. Mahdi, C. Martin-Barreiro, and X. Cabezas, “A Novel Hybrid Approach Using an Attention-Based Transformer + GRU Model for Predicting Cryptocurrency Prices,”arXiv preprint arXiv:2504.17079, 2025

    2025 arXiv

  47. [61]

    Online Deep Learning for Real-Time Stock Trend Prediction,

    Y . Deng, X. Zhou, Q. Zhao, and X. Li, “Online Deep Learning for Real-Time Stock Trend Prediction,”Applied Sciences, vol. 11, no. 12, p. 5446, 2021. [Online]. Available: https://doi.org/10.3390/app11125446

    2021 doi

  48. [62]

    Crypto-News Sentiment Analysis Using LSTM with Attention Mechanism,

    I. Sattarov and S. Lee, “Crypto-News Sentiment Analysis Using LSTM with Attention Mechanism,”Entropy, vol. 24, no. 9, p. 1279, 2022. [Online]. Available: https://doi.org/10.3390/e24091279

    2022 doi

  49. [63]

    Hybrid Forecasting Model Using Transformer and XGBoost for Time Series Prediction,

    Z. Wang and R. Jin, “Hybrid Forecasting Model Using Transformer and XGBoost for Time Series Prediction,”Neural Computing and Applications, vol. 35, pp. 12791–12808, 2023. [Online]. Available: https://doi.org/10.1007/s00521-023-08083-5

    2023 doi

  50. [64]

    Predicting Cryptocurrency Prices Using Sentiment and Technical Indicators,

    S. Tiwari, S. Saha, and R. Dash, “Predicting Cryptocurrency Prices Using Sentiment and Technical Indicators,”IEEE Access, vol. 9, pp. 37207–37219, 2021. [Online]. Available: https://doi.org/10.1109/ ACCESS.2021.3063710

    2021

  51. [65]

    Time-Series Forecasting Using SVMD- LSTM: A Hybrid Approach for Stock Market Prediction,

    A. Agarwal and R. Sharma, “Time-Series Forecasting Using SVMD- LSTM: A Hybrid Approach for Stock Market Prediction,”Journal of Probability and Statistics, vol. 2025, Article ID 9464938, 2025

    2025

  52. [66]

    LSTM Based Sentiment Analysis for Cryptocurrency Prediction,

    X. Huang, W. Zhang, X. Tang, M. Zhang, J. Surbiryala, V . Iosifidis, and Z. Liu, “LSTM Based Sentiment Analysis for Cryptocurrency Prediction,”arXiv preprint arXiv:2103.14804, 2021

    2021 arXiv

  53. [67]

    Transfer Learning for Cross-Market Stock Price Prediction,

    W. Zhang, X. Zeng, and Y . Liu, “Transfer Learning for Cross-Market Stock Price Prediction,”Pattern Recognition Letters, vol. 143, pp. 88–95,

  54. [68]

    Combining Multivariate Time Series and Graph Neural Networks for Financial Forecasting,

    H. Yu, S. Wu, and F. Wang, “Combining Multivariate Time Series and Graph Neural Networks for Financial Forecasting,”Information Sciences, vol. 590, pp. 203–217, 2022. [Online]. Available: https://doi. org/10.1016/j.ins.2022.01.025

    2022 doi

  55. [69]

    Real-Time Deep Learning Pipeline for Cryp- tocurrency Price Prediction on Edge Devices,

    A. Ahmed and L. Zhao, “Real-Time Deep Learning Pipeline for Cryp- tocurrency Price Prediction on Edge Devices,”Sensors, vol. 25, no. 2, p. 467, 2025. [Online]. Available: https://doi.org/10.3390/s25020467

    2025 doi

  56. [70]

    A Survey on Explainable AI for Deep Learning-Based Time Series Forecasting,

    M. R. Islam and A. Rahman, “A Survey on Explainable AI for Deep Learning-Based Time Series Forecasting,”Artificial Intelligence Review, vol. 56, pp. 3751–3791, 2023. [Online]. Available: https://doi.org/10. 1007/s10462-023-10483-4

    2023

  57. [71]

    SHAP-Based Explanation of XGBoost-LSTM Hy- brid Models in Financial Forecasting,

    Y . Hong and Y . Yin, “SHAP-Based Explanation of XGBoost-LSTM Hy- brid Models in Financial Forecasting,”Expert Systems with Applications, vol. 234, p. 120395, 2024. [Online]. Available: https://doi.org/10.1016/ j.eswa.2023.120395

    2024

  58. [72]

    Multi-objective Evolutionary Optimization in Cryptocurrency Portfolio Design,

    S. Sarwar and T. Ahmed, “Multi-objective Evolutionary Optimization in Cryptocurrency Portfolio Design,”Applied Soft Computing, vol. 120, p. 108668, 2022. [Online]. Available: https://doi.org/10.1016/j.asoc.2022. 108668

    2022 doi

  59. [74]

    Available: https://doi.org/10.1016/j.patrec.2020.12.002

    [Online]. Available: https://doi.org/10.1016/j.patrec.2020.12.002

    2020 doi

  60. [2016]

    https://doi.org/10.1080/00036846.2015.1109038

    2015

  61. [2018]

    https://doi.org/10.1371/journal.pone.0195200

    doi

  62. [2019]

    https://doi.org/10.3390/math7100898

    doi

  63. [2021]

    https://doi.org/10.1109/ACCESS.2021.3099882

    2021

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.