Pith. sign in

REVIEW 4 major objections 6 minor 44 references

Boosting the Accuracy of Stock Market Prediction via Multi-Layer Hybrid MTL Structure

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read This paper proposes a multi-layer hybrid multi-task architecture, combining a Transformer encoder, a KAN layer, and a bidirectional GRU, and reports $R^2$ up to 0.98 along with lower MAE and RMSE than the compared models.

desk verdict Incremental architecture stacked from published parts; the reported results cannot be believed because the paper's own numbers conflict by orders of magnitude and the dataset is never named. read the letter →

arxiv 2501.09760 v1 pith:LVUOCLTA submitted 2025-01-01 q-fin.ST cs.LG

Yuxi Hong This is my paper · ORCID
classification q-fin.STcs.LG
keywords stockmarketpredictionmulti-tasklearningTransformerencoderBiGRUKolmogorov-Arnoldnetworkhybriddeeptimeseriesforecasting
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 proposes a multi-layer hybrid multi-task learning (MTL) architecture for stock-price forecasting. The model stacks a Transformer encoder, a Kolmogorov-Arnold Network (KAN) layer, and a bidirectional GRU (BiGRU), then trains them jointly so that several market features are predicted from shared representations. The reported experiments claim that this combination beats each component model, each paired combination, and four published baselines, with $R^2$ as high as 0.98 and lower MAE, RMSE, and MAPE. The central claim is that combining these three learning mechanisms in one multi-task network yields more accurate and stable stock forecasts than any single block, and that this generalizes to noisy, nonlinear financial time series.

What carries the argument

The carrying mechanism is the serial stack of three learned blocks. The Transformer encoder uses multi-head self-attention to identify complex correspondences across the input feature channels; the KAN layer replaces fixed node activations with learned univariate edge functions, giving flexible nonlinear approximation; and the BiGRU processes the sequence in both directions, producing a hidden state that concatenates forward and backward context and is fed to a dense output layer. The multi-task formulation is the fourth piece: a shared latent representation is used to predict several output features simultaneously, which the paper argues is the source of joint accuracy gains.

What would settle it

Retrain the same architecture with the BiGRU made causal or replaced by a unidirectional GRU, keeping all other settings and the chronological data split unchanged; if the $R^2$ drops substantially on the same test period, the headline fit comes from look-ahead rather than from predictive skill.

Watch

Extended reading notes

Core claim

The paper's central claim is that a specific three-layer stack — Transformer encoder, KAN, and BiGRU — trained with a multi-task objective gives materially better stock forecasts than the components used alone or in pairs. The reported support is a comparison against four published methods and five ablation configurations: the proposed structure reaches an average $R^2$ of 0.977 in the advanced-method table, 0.968 in the ablation table, and up to 0.98 in the abstract; its MAE and RMSE are likewise lower in both tables. The mechanism offered is that the Transformer's self-attention extracts complex correspondences among input features, the KAN layer approximates nonlinear functions through learned edge functions, and the BiGRU reads the sequence in both directions so that each prediction is informed by the full input window. Multi-task sharing of the learned representations is presented as the reason the joint predictions of trading volume and trading amount improve together.

Load-bearing premise

The headline accuracy rests on treating the bidirectional GRU's ability to read both past and future parts of the input window as a legitimate way to test a stock-price forecaster; if a real forecast cannot use values beyond the point being predicted, the reported $R^2$ is not a valid out-of-sample result.

Editorial extensions

If this is right

  • On the paper's reported numbers, the hybrid MTL structure achieves the lowest MAE, RMSE, and MAPE and the highest $R^2$ among all compared advanced methods and ablation models.
  • The multi-task sharing of latent representations becomes a workable way to improve per-feature prediction, since trading volume and trading amount are predicted jointly from the same learned features.
  • The reported inference time of 0.0193 positions the architecture as usable in near-real-time prediction, not just offline backtesting.
  • The structure can serve as an ensemble component: the paper explicitly notes that combining it with other forecasters is a route to further accuracy gains.

Reading between the lines

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

  • If the bidirectional GRU literally uses future tokens inside each prediction window, the high $R^2$ values are more plausibly in-window interpolation than out-of-sample forecasting; the causal retraining in the falsifier would settle this.
  • The architecture is generic enough to transfer to other high-dimensional, noisy time-series problems such as electricity load or traffic flow, but the paper reports only stock data and does not test this.
  • The abstract, tables, and conclusion report MAE figures of 1.078, 16.0–18.4, and 0.45 and $R^2$ figures of 0.98, 0.977/0.968; reconciling these would be needed before any single headline number is treated as the true size of the gain.
  • The comparison against a hybrid BiLSTM-GRU baseline and the ablation results suggest that the KAN layer and the multi-task sharing, rather than the Transformer alone, drive the reported improvement; a one-at-a-time ablation would make this attribution explicit.
Share X Bluesky LinkedIn Reddit HN

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 stock-price forecasting architecture that combines a Transformer encoder, a bidirectional GRU (BiGRU), and a Kolmogorov-Arnold Network (KAN) within a multi-task learning (MTL) framework, and reports MAE, RMSE, MAPE, and R² on comparisons with several deep-learning baselines and ablation variants. The abstract claims an MAE as low as 1.078, MAPE as low as 0.012, and R² as high as 0.98, with further claims that the method outperforms all compared models. The paper includes an architecture table, equations for the attention mechanism, KAN, and BiGRU, and an experimental section with two comparison tables, figures, and a short inference-time table.

Significance. If the reported results were reproducible, the proposed three-component hybrid would be a useful contribution to the stock-forecasting literature, and the combination of attention, bidirectional recurrent processing, and KAN layers is a plausible design direction. However, the manuscript as submitted does not establish the central claim: the reported metrics are internally inconsistent across the abstract, the conclusion, and Tables 2–3, the tables appear to be swapped relative to the prose, and essential experimental details such as the dataset, split, and target definition are absent. The paper provides no code and no data, so the numerical evidence cannot be independently verified. The idea has some interest, but the evidence in the manuscript is not sufficient to support the claimed state-of-the-art performance.

major comments (4)
  1. [Abstract; Section 4.3; Section 4.4; Section 5; Tables 2–3] The central reported metrics are mutually inconsistent. The abstract reports MAE as low as 1.078 and R² as high as 0.98; the conclusion reports MAE as low as 0.45 and R² up to 0.98; Table 3 reports MAE 16.007 and R² 0.977 for the proposed method; Table 2 reports MAE 18.410 and R² 0.968; and Section 4.4 states that cross-validation gives an average test R² of 0.9831 that appears in no table. These numbers are not compatible with one another and cannot all describe the same average experimental result. The authors must state which numbers correspond to which experiment, dataset, and evaluation setting, and must correct the inconsistencies before the central claim can be assessed.
  2. [Sections 4.3 and 4.4; Tables 2 and 3] The comparison tables are swapped relative to the prose. Section 4.3 claims to compare against advanced methods by Seo et al., Ma et al., Gao et al., and Hemajothi et al., but Table 2 lists only ablation variants (KAN, Transformer, BiGRU, KAN-Transformer, Transformer-BiGRU, Proposed). Section 4.4 claims to analyze ablation methods, but Table 3 lists the four external methods and the proposed method. The text in Section 4.3 quotes RMSE 39.820, which is the Table 3 value, while Section 4.4 quotes RMSE 21.004, which is the Table 2 value. This mismatch makes it impossible to read the empirical comparison as currently written.
  3. [Section 4.1; Table 1] The experimental setup is not reproducible. The paper never names the dataset, the exchange, the tickers, the date range, the train/validation/test split, the scaling procedure, the prediction horizon, or the number of repeated runs. Table 1 lists an input shape of (None, 5, 1), which contradicts the text's claim of six input features (Open, Close, High, Low, Volume, Amount) and the stated goal of predicting multiple target features over multiple future time steps; the final layer is also a single output (None, 1). Additionally, the phrase 'multi-task learning' is never operationalized: there is no description of the task heads, the loss function, or the task weighting. Without these details, the reported numerical results cannot be verified.
  4. [Section 3.5, Eq. (8)] The right-to-left GRU in Eq. (8) is not by itself proof of look-ahead leakage, because with a 5-step input window and a single next-step output the backward pass only sees positions inside the window, which are available at prediction time in a sliding-window setup. However, the manuscript never defines the temporal alignment between the input window and the target. If the target is the next observation after the window, the BiGRU is fine; if the target is an observation inside the window, or if the evaluation is performed as in-window reconstruction, then Eq. (8) leaks target information. The sentence in Section 3.2 that the BiGRU 'ensures that both past and future contexts are considered simultaneously' must be reconciled with a genuine forecasting setup, and the exact prediction target and horizon must be stated explicitly.
minor comments (6)
  1. [Eq. (10)] Equation (10) applies a softmax activation to the regression output y_t; this is not a standard choice for continuous price prediction and appears inconsistent with the MAE/RMSE regression metrics used in Section 4.2.
  2. [Eq. (14)] The R² formula is written as R2 = I - SSresidual/SStotal; the symbol 'I' should be '1' to denote the constant one.
  3. [Figures 4–13; Section 4.4] The captions of Figures 4–13 are all identical to the caption of Figure 1 and do not describe the displayed curves; Section 4.4 also references Fig. 11 twice and says 'four representative comparison results' while listing five figures. These need correction.
  4. [Eqs. (11)–(13)] The variables in the MAE and MAPE formulas are inconsistent: Eq. (11) uses y_i for predicted and ¯y_i for actual, while Eq. (13) uses y_i in the denominator as the actual value. The predicted and true values should be defined once and used consistently.
  5. [Table 1] The parameter counts for the Transformer layer (4 parameters) and KAN layer (2 parameters) are implausibly small for the described operations; the authors should verify and explain the parameter accounting.
  6. [Throughout] There are numerous typographical errors, including 'diffrent', 'perdition', 'Forthermore', and 'sof tmax', and the notation in the problem statement of Section 3.1 (e.g., a_{mn+1}) is not clearly defined. A careful proofread is needed.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the reported metric inconsistencies, missing dataset details, and potential look-ahead are correctness and reproducibility concerns, not reductions of the prediction to the paper's own inputs.

full rationale

The paper's derivation chain is a standard supervised sequence model: an input window is transformed by Transformer, KAN, and BiGRU layers and mapped to a single dense output. No equation defines a target in terms of a fitted parameter drawn from that same target, and no load-bearing argument relies on a self-citation chain or on an imported uniqueness theorem. The BiGRU equations (7)-(9) describe forward and backward hidden states over the input sequence; although the right-to-left pass means each position can see later positions within the input window, the paper's problem statement places the targets after the input window, so the text does not exhibit a by-construction equivalence between prediction and fit. The serious inconsistencies among the reported headline numbers (Abstract MAE 1.078 and R2 0.98; Conclusion MAE 0.45; Table 2 MAE 18.410; Table 3 MAE 16.007; Section 4.4 average test R2 0.9831) and the absence of a named dataset, split, or scaling procedure make the central empirical claim unverifiable, but those are reproducibility and correctness defects rather than circularity under the enumerated patterns. The paper is therefore not circular in the sense of deriving its results from its own assumptions: it simply does not provide enough verifiable empirical evidence to support those results.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The model composition uses only published components and introduces no new theoretical object. The central empirical claim rests on unstated dataset details, hand-chosen hyperparameters, and a bidirectional look-ahead setup that contaminates the evaluation.

free parameters (4)
  • Number of attention heads = 4
    Chosen by hand in Section 3.3; no sensitivity analysis is provided.
  • Hidden dimensions = 256 (BiGRU), 32 (Transformer)
    Architecture choices from Table 1; the paper does not justify or tune these values.
  • Sequence length = 5
    Input shape is (None, 5, 1) in Table 1; no reason is given for using five time steps.
  • Dropout rate = Not stated
    Dropout layers appear in Table 1, but the rate is not specified.
assumptions (3)
  • domain assumption Stock price time series contain predictable structure from OHLCV features
    The entire forecasting task in Section 3.1 assumes predictive signal exists in the chosen features, a contested assumption in finance.
  • domain assumption The evaluation dataset is representative and splits avoid leakage
    No dataset is described in Section 4.1, yet the success of the method depends on the unstated representativeness and clean splitting of the data.
  • ad hoc to paper Bidirectional context is allowed during prediction
    Section 3.5 uses future information without masking, which is invalid for genuine forecasting and leaks the target into the features.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Boosting the Accuracy of Stock Market Prediction via Multi-Layer Hybrid MTL Structure." pith.science (2026). https://pith.science/paper/LVUOCLTA

@misc{pith2026250109760,
  author       = {Pith},
  title        = {Pith review of: Boosting the Accuracy of Stock Market Prediction via Multi-Layer Hybrid MTL Structure},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LVUOCLTA}},
  note         = {Machine review of arXiv:2501.09760}
}
read the original abstract

Accurate stock market prediction provides great opportunities for informed decision-making, yet existing methods struggle with financial data's non-linear, high-dimensional, and volatile characteristics. Advanced predictive models are needed to effectively address these complexities. This paper proposes a novel multi-layer hybrid multi-task learning (MTL) framework aimed at achieving more efficient stock market predictions. It involves a Transformer encoder to extract complex correspondences between various input features, a Bidirectional Gated Recurrent Unit (BiGRU) to capture long-term temporal relationships, and a Kolmogorov-Arnold Network (KAN) to enhance the learning process. Experimental evaluations indicate that the proposed learning structure achieves great performance, with an MAE as low as 1.078, a MAPE as low as 0.012, and an R^2 as high as 0.98, when compared with other competitive networks.

Figures

Figures reproduced from arXiv: 2501.09760 by the authors.

Figure 1
Figure 1. The multi-layer hybrid MTL structure for stock market prediction, involving an enhanced Transformer, KAN and BiGRU. [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. The multi-head attention in the Transformer structure. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. The detailed comparison of MLP and KAN structure [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: The multi-layer hybrid MTL structure for stock market prediction, involving an enhanced Transformer, KAN and BiGRU. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: The multi-layer hybrid MTL structure for stock market prediction, involving an enhanced Transformer, KAN and BiGRU. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: The multi-layer hybrid MTL structure for stock market prediction, involving an enhanced Transformer, KAN and BiGRU. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: The multi-layer hybrid MTL structure for stock market prediction, involving an enhanced Transformer, KAN and BiGRU. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: The multi-layer hybrid MTL structure for stock market prediction, involving an enhanced Transformer, KAN and BiGRU. [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: The multi-layer hybrid MTL structure for stock market prediction, involving an enhanced Transformer, KAN and BiGRU. [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: The multi-layer hybrid MTL structure for stock market prediction, involving an enhanced Transformer, KAN and BiGRU. [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 11
Figure 11. Figure 11: The multi-layer hybrid MTL structure for stock market prediction, involving an enhanced Transformer, KAN and BiGRU. [PITH_FULL_IMAGE:figures/full_fig_p010_11.png]
Figure 12
Figure 12. Figure 12: The multi-layer hybrid MTL structure for stock market prediction, involving an enhanced Transformer, KAN and BiGRU. [PITH_FULL_IMAGE:figures/full_fig_p011_12.png]
Figure 13
Figure 13. Figure 13: The multi-layer hybrid MTL structure for stock market prediction, involving an enhanced Transformer, KAN and BiGRU. [PITH_FULL_IMAGE:figures/full_fig_p011_13.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 41 canonical work pages

  1. [1]

    A stock market trends analysis of reliance using machine learn- ing techniques

    Vineet Kukreti, Chandradeep Bhatt, and Rakesh Dani. A stock market trends analysis of reliance using machine learn- ing techniques. In 2023 6th International Conference on In- formation Systems and Computer Networks (ISCON) , pages 9 Figure 10. The multi-layer hybrid MTL structure for stock market prediction, involving an enhanced Transformer, KAN and BiG...

    work page 2023

  2. [2]

    Progress and prospects of data-driven stock price fore- casting research

    Chuanjun Zhao, Meiling Wu, Jingfeng Liu, Zening Duan, Li- hua Shen, Xuekui Shangguan, Donghang Liu, Yanjie Wang, et al. Progress and prospects of data-driven stock price fore- casting research. International Journal of Cognitive Com- puting in Engineering, 4:100–108, 2023. 1

    work page 2023

  3. [3]

    A study of the impact of covid-19 on the chinese stock market based on a new textual multiple arma model

    Weijun Xu, Zhineng Fu, Hongyi Li, Jinglong Huang, Wei- dong Xu, and Yiyang Luo. A study of the impact of covid-19 on the chinese stock market based on a new textual multiple arma model. Statistical Analysis and Data Mining: The ASA Data Science Journal, 16(1):5–15, 2023. 1

    work page 2023

  4. [4]

    Stock price prediction using the arima model

    Adebiyi A Ariyo, Adewumi O Adewumi, and Charles K Ayo. Stock price prediction using the arima model. In 2014 UKSim-AMSS 16th international conference on com- puter modelling and simulation, pages 106–112. IEEE, 2014. 1

    work page 2014

  5. [5]

    Stock return prediction under garch—an empirical assessment

    Helmut Herwartz. Stock return prediction under garch—an empirical assessment. International Journal of Forecasting, 33(3):569–580, 2017. 1

    work page 2017

  6. [6]

    Fore- casting indonesian stock index using arma-garch model

    Dwi Susanti, Kirana Fara Labitta, and Sukono Sukono. Fore- casting indonesian stock index using arma-garch model. In- ternational Journal of Quantitative Research and Modeling, 5(2):199–204, 2024. 1

    work page 2024

  7. [7]

    Comparison of arima, ann and lstm for stock price prediction

    Qihang Ma. Comparison of arima, ann and lstm for stock price prediction. In E3S Web of Conferences, volume 218, page 01026. EDP Sciences, 2020. 1

    work page 2020

  8. [8]

    China’s com- mercial bank stock price prediction using a novel k-means- lstm hybrid approach

    Yufeng Chen, Jinwang Wu, and Zhongrui Wu. China’s com- mercial bank stock price prediction using a novel k-means- lstm hybrid approach. Expert Systems with Applications , 202:117370, 2022. 1

    work page 2022

Show all 44 references
  1. [9]

    A new cnn-based model for financial time series: Taiex and ftse stocks fore- casting

    Melih Kirisci and Ozge Cagcag Yolcu. A new cnn-based model for financial time series: Taiex and ftse stocks fore- casting. Neural Processing Letters, 54(4):3357–3374, 2022. 1

    2022

  2. [10]

    Stock price prediction using rnns: A 10 Figure 12

    Sahil Agarwal, Bosco Paul Alapatt, Akhil M Nair, and Fr Jossy George. Stock price prediction using rnns: A 10 Figure 12. The multi-layer hybrid MTL structure for stock market prediction, involving an enhanced Transformer, KAN and BiGRU. Figure 13. The multi-layer hybrid MTL st...

    2024

  3. [11]

    A graph-based cnn-lstm stock price prediction algorithm with leading indicators

    Jimmy Ming-Tai Wu, Zhongcui Li, Norbert Herencsar, Bay V o, and Jerry Chun-Wei Lin. A graph-based cnn-lstm stock price prediction algorithm with leading indicators. Multime- dia Systems, 29(3):1751–1770, 2023. 1

    2023

  4. [12]

    Graph-based stock correlation and prediction for high-frequency trading systems

    Tao Yin, Chenzhengyi Liu, Fangyu Ding, Ziming Feng, Bo Yuan, and Ning Zhang. Graph-based stock correlation and prediction for high-frequency trading systems. Pattern Recognition, 122:108209, 2022. 1

    2022

  5. [13]

    Inte- grated gcn-lstm stock prices movement prediction based on knowledge-incorporated graphs construction

    Yong Shi, Yunong Wang, Yi Qu, and Zhensong Chen. Inte- grated gcn-lstm stock prices movement prediction based on knowledge-incorporated graphs construction. International Journal of Machine Learning and Cybernetics , 15(1):161– 176, 2024. 1

    2024

  6. [14]

    Stock market index prediction using transformer neural network models and frequency decomposition

    Camilo Ya ˜nez, Werner Kristjanpoller, and Marcel C Minu- tolo. Stock market index prediction using transformer neural network models and frequency decomposition. Neural Com- puting and Applications, pages 1–21, 2024. 1

    2024

  7. [15]

    Data-driven stock forecasting models based on neural networks: A review

    Wuzhida Bao, Yuting Cao, Yin Yang, Hangjun Che, Jun- jian Huang, and Shiping Wen. Data-driven stock forecasting models based on neural networks: A review. Information Fusion, page 102616, 2024. 1

    2024

  8. [16]

    Achieving more with less: A tensor-optimization-powered ensemble method

    Jinghui Yuan, Weijin Jiang, Zhe Cao, Fangyuan Xie, Rong Wang, Feiping Nie, and Yuan Yuan. Achieving more with less: A tensor-optimization-powered ensemble method. arXiv preprint arXiv:2408.02936, 2024. 2

    2024 arXiv

  9. [17]

    A margin-maximizing fine-grained ensemble method

    Jinghui Yuan, Hao Chen, Renwei Luo, and Feiping Nie. A margin-maximizing fine-grained ensemble method. arXiv preprint arXiv:2409.12849, 2024. 2

    2024 arXiv

  10. [18]

    A survey on ensemble learning

    Xibin Dong, Zhiwen Yu, Wenming Cao, Yifan Shi, and Qianli Ma. A survey on ensemble learning. Frontiers of Computer Science, 14:241–258, 2020. 2

    2020

  11. [19]

    Application of 11 an ann and lstm-based ensemble model for stock market pre- diction

    Fang Liu, Shaobo Guo, Qianwen Xing, Xinye Sha, Ying Chen, Yuhui Jin, Qi Zheng, and Chang Yu. Application of 11 an ann and lstm-based ensemble model for stock market pre- diction. In 2024 IEEE 7th International Conference on Infor- mation Systems and Computer Aided Education (I...

    2024

  12. [20]

    Stock market prediction with time series data and news headlines: a stacking ensem- ble approach

    Roberto Corizzo and Jacob Rosen. Stock market prediction with time series data and news headlines: a stacking ensem- ble approach. Journal of Intelligent Information Systems , 62(1):27–56, 2024. 2

    2024

  13. [21]

    An effective time series analysis for stock trend prediction using arima model for nifty midcap-50

    B Uma Devi, Darshan Sundar, and P Alli. An effective time series analysis for stock trend prediction using arima model for nifty midcap-50. International Journal of Data Mining & Knowledge Management Process, 3(1):65, 2013. 2

    2013

  14. [22]

    Stock price pre- diction using arima model

    Supriti Khanderwal and Debasis Mohanty. Stock price pre- diction using arima model. International Journal of Market- ing & Human Resource Research, 2(2):98–107, 2021. 2

    2021

  15. [23]

    Prediction of popular global stock in- dexes volatility by using arch/garch models.GARCH Models (July 24, 2024), 2024

    Nagendra Marisetty. Prediction of popular global stock in- dexes volatility by using arch/garch models.GARCH Models (July 24, 2024), 2024. 2

    2024

  16. [24]

    Stock market prediction using hidden markov models

    Aditya Gupta and Bhuwan Dhingra. Stock market prediction using hidden markov models. In 2012 Students Conference on Engineering and Systems, pages 1–4. IEEE, 2012. 2

    2012

  17. [25]

    Research on hmm-based efficient stock price prediction

    Zhi Su and Bo Yi. Research on hmm-based efficient stock price prediction. Mobile Information Systems , 2022(1):8124149, 2022. 2

    2022

  18. [26]

    Network log-arch models for forecasting stock market volatility

    Raffaele Mattera and Philipp Otto. Network log-arch models for forecasting stock market volatility. International Journal of Forecasting, 2024. 2

    2024

  19. [27]

    Stock price prediction: a comparative study between traditional statisti- cal approach and machine learning approach

    Indronil Bhattacharjee and Pryonti Bhattacharja. Stock price prediction: a comparative study between traditional statisti- cal approach and machine learning approach. In 2019 4th international conference on electrical information and com- munication technology (EICT), pages 1–...

    2019

  20. [28]

    Stock closing price prediction using machine learning techniques

    Mehar Vijh, Deeksha Chandola, Vinay Anand Tikkiwal, and Arun Kumar. Stock closing price prediction using machine learning techniques. Procedia computer science, 167:599– 606, 2020. 2

    2020

  21. [29]

    Stock price prediction using machine learning and lstm-based deep learning models

    Sidra Mehtab, Jaydip Sen, and Abhishek Dutta. Stock price prediction using machine learning and lstm-based deep learning models. In Machine Learning and Metaheuristics Algorithms, and Applications: Second Symposium, SoMMA 2020, Chennai, India, October 14–17, 2020, Revised Se- ...

    2020

  22. [30]

    Multivariate and multi-task deep learning architectures for improved stock market prediction and risk management

    Osama Assaf. Multivariate and multi-task deep learning architectures for improved stock market prediction and risk management. PhD thesis, University of Reading, 2024. 3

    2024

  23. [31]

    Stock movement prediction: A multi-input lstm approach

    Pan Tang, Cheng Tang, and Keren Wang. Stock movement prediction: A multi-input lstm approach. Journal of Fore- casting, 2024. 3

    2024

  24. [32]

    A multifactor analysis model for stock market prediction

    Akash Deep. A multifactor analysis model for stock market prediction. International Journal of Computer Science and Telecommunications, 14(1), 2023. 3

    2023

  25. [33]

    Hybrid intelligent systems for stock market analysis

    Ajith Abraham, Baikunth Nath, and Prabhat Kumar Mahanti. Hybrid intelligent systems for stock market analysis. In Computational Science-ICCS 2001: International Confer- ence San Francisco, CA, USA, May 28—30, 2001 Proceed- ings, Part II 1, pages 337–345. Springer, 2001. 3

    2001

  26. [34]

    Stock mar- ket analysis: A review and taxonomy of prediction tech- niques

    Dev Shah, Haruna Isah, and Farhana Zulkernine. Stock mar- ket analysis: A review and taxonomy of prediction tech- niques. International Journal of Financial Studies, 7(2):26,

  27. [35]

    Stock price prediction based on arima-rnn combined model

    YU Shui-Ling and Zhe Li. Stock price prediction based on arima-rnn combined model. In 4th International Conference on Social Science (ICSS 2017), pages 1–6, 2017. 3

    2017

  28. [36]

    Research on stock index prediction based on arima-cnn-lstm model

    Ziyan Zhang. Research on stock index prediction based on arima-cnn-lstm model. In 9th International Conference on Financial Innovation and Economic Development (ICFIED 2024), pages 558–565. Atlantis Press, 2024. 3

    2024

  29. [37]

    Stock price prediction based on lstm and lightgbm hybrid model

    Liwei Tian, Li Feng, Lei Yang, and Yuankai Guo. Stock price prediction based on lstm and lightgbm hybrid model. The Journal of Supercomputing, 78(9):11768–11793, 2022. 3

    2022

  30. [38]

    Modal decomposition-based hybrid model for stock index predic- tion

    Pin Lv, Yating Shu, Jia Xu, and Qinjuan Wu. Modal decomposition-based hybrid model for stock index predic- tion. Expert Systems with Applications , 202:117252, 2022. 3

    2022

  31. [39]

    Kan: Kolmogorov-arnold networks

    Ziming Liu, Yixuan Wang, Sachin Vaidya, Fabian Ruehle, James Halverson, Marin Solja ˇci´c, Thomas Y Hou, and Max Tegmark. Kan: Kolmogorov-arnold networks. arXiv preprint arXiv:2404.19756, 2024. 4

    2024 arXiv

  32. [40]

    Kolmogorov-arnold networks (kans) for time series analysis

    Cristian J Vaca-Rubio, Luis Blanco, Roberto Pereira, and M`arius Caus. Kolmogorov-arnold networks (kans) for time series analysis. arXiv preprint arXiv:2405.08790, 2024. 4

    2024

  33. [41]

    Exploring the efficient market hypothesis for accurate stock movement prediction via feature-axis transformer

    Kanghyeon Seo and Jihoon Yang. Exploring the efficient market hypothesis for accurate stock movement prediction via feature-axis transformer. In Proceedings of the 39th ACM/SIGAPP Symposium on Applied Computing , pages 892–901, 2024. 6, 8, 9

    2024

  34. [42]

    Predicting stock prices using hybrid lstm and arima model

    Chi Ma, Jie Wu, Hui Hu, Yue Nai Chen, and Jing Yan Li. Predicting stock prices using hybrid lstm and arima model. IAENG International Journal of Applied Mathemat- ics, 54(3), 2024. 6, 8, 9

    2024

  35. [43]

    Attention based hybrid parametric and neural network models for non-stationary time series prediction

    Zidi Gao and Ercan Engin Kuruo ˘glu. Attention based hybrid parametric and neural network models for non-stationary time series prediction. Expert Systems, 41(2):e13419, 2024. 6, 8, 9

    2024

  36. [44]

    Prediction of stock price model through the implementation of hybrid bilstm-gru

    S Hemajothi, S Sivasaravana Babu, V Prathiksha, B Sangeetha, E Sivani, and K Soundari. Prediction of stock price model through the implementation of hybrid bilstm-gru. In 2024 International Conference on Recent Advances in Electrical, Electronics, Ubiquitous Communi- cation, a...

    2024

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.