Paper: arXiv 2502.15853

Authors: Daksh Dave, Gauransh Sawhney, Vikhyat Chauhan

Abstract

This paper presents a comprehensive study on stock price prediction, leveragingadvanced machine learning (ML) and deep learning (DL) techniques to improve financial forecasting accuracy. The research evaluates the performance of various recurrent neural network (RNN) architectures, including Long Short-Term Memory (LSTM) networks, Gated Recurrent Units (GRU), and attention-based models. These models are assessed for their ability to capture complex temporal dependencies inherent in stock market data. Our findings show that attention-based models outperform other architectures, achieving the highest accuracy by capturing both short and long-term dependencies. This study contributes valuable insights into AI-driven financial forecasting, offering practical guidance for developing more accurate and efficient trading systems.

Complexity vs Empirical Score

  • Math Complexity: 7.0/10
  • Empirical Rigor: 6.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper demonstrates high mathematical complexity through detailed exposition of LSTM/GRU/Transformer architectures and their gating mechanisms, while empirical rigor is elevated by its use of specific real-world datasets (Tesla stock), comparison of multiple architectures (RNNs, CNNs, Transformers), and focus on practical trading system applications.

Research Flowchart

  flowchart TD
  A["Research Goal: Enhance Stock Price Prediction Accuracy<br>using AI/ML"] --> B["Methodology: Evaluate ML/DL Models<br>on Historical Stock Data"]
  B --> C["Data Inputs: Historical Stock Prices<br>Time-Series Data"]
  C --> D["Computational Process: Train & Compare<br>LSTM, GRU, & Attention-Based Models"]
  D --> E["Key Finding: Attention-Based Models<br>Outperform (Highest Accuracy)"]
  E --> F["Outcome: Improved Financial Forecasting<br>for Real-Time Trading Strategies"]

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 a comprehensive evaluation of recurrent neural network architectures for stock price prediction, highlighting attention-based models as the top-performing approach.

Source sentence
Our findings show that attention-based models outperform other architectures, achieving the highest accuracy by capturing both short and long-term dependencies.

Assumes 4

  • The model is trained on historical data up to 30 days before the final timestamp and tested on the last 30 days. data
    Source sentence
    The training set comprises all data points from the initial timestamp up to 30 days prior to the final recorded timestamp. The test set includes the data from the last 30 days of the dataset.
  • Each model uses a single hidden layer of 128 units and is trained for 10 epochs with a learning rate of 0.01. model
    Source sentence
    For each model in the experiments, the number of layers was set to 1. The size of the hidden layer was fixed at 128 units. Training was performed over 10 epochs and the learning rate was set to 0.01.
  • The Tesla stock dataset contains standard OHLCV fields and is representative of market price dynamics. data
    Source sentence
    A Tesla stock dataset typically contains historical data on the stock performance of Tesla Inc. (TSLA) over a given period of time.
  • The experiments are conducted using Python on Google Colab with L4 GPUs and TensorFlow. execution
    Source sentence
    For the implementation, we use Python in a Jupyter notebook environment hosted on Google Colab and use L4 GPUs for training. The TensorFlow framework is employed to build and trai…

Data

  • Tesla Stock dataset market-prices unknown
    Tesla Inc. (TSLA)
    Source sentence
    ... Date Open High Low Close Adj Close Volume 2024-01-02 720.00 725.00 710.00 715.00 715.00 15,200,000

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • Attention-based models achieve the highest predictive accuracy among the evaluated RNN architectures. predictive-accuracy
    Source sentence
    Our findings show that attention-based models outperform other architectures, achieving the highest accuracy by capturing both short and long-term dependencies.
  • GRU models generally lag behind LSTM variants, with GRU-Seq2Seq reaching 90.8855% accuracy. performance
    Source sentence
    GRU-Seq2Seq (Figure 11) improved significantly to 90.8855%.
  • Bidirectional and Seq2Seq enhancements improve baseline LSTM and GRU performance. performance
    Source sentence
    Bidirectional processing enriches the model’s contextual understanding by incorporating information from both past and future time steps.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 36 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 (48)

  1. Stock market prediction using machine learning (2018)
  2. A survey on stock market prediction using machine learning techniques (2020)
  3. Stock price prediction using news sentiment analysis (2019)
  4. Combine facebook prophet and LSTM with BPNN forecasting financial markets: the morgan Taiwan index (2019)
  5. Stock market prediction using machine learning techniques (2019)
  6. Applying machine learning models in stock market prediction (2018)
  7. Novel deep learning model with CNN and bi-directional LSTM for improved stock market index prediction (2019)
  8. Indian stock market prediction using artificial neural networks on tick data (2019)
and 40 more
  1. Stock price prediction using DEEP learning algorithm and its comparison with machine learning algorithms (2019)
  2. Stock price movements classification using machine and deep learning techniques-the case study of indian stock market (2019)
  3. Predicting stock market trends using machine learning algorithms via public sentiment and political situation analysis (2020)
  4. Predicting stock market trends using machine learning and deep learning algorithms via continuous and binary data; a comparative analysis (2020)
  5. Stock market prediction based on social sentiments using machine learning (2018)
  6. Stock market prediction using machine learning classifiers and social media, news (2022)
  7. Proceedings of the Genetic and Evolutionary Computation Conference (GECCO 2018) (2018) — link
  8. Stock market prediction using LSTM recurrent neural network (2020)
  9. Real-time Statistical Log Anomaly Detection with Continuous AIOps Learning. (2022)
  10. Hybrid machine learning to detect anomalies (2022)
  11. Fixme: Enhance software reliability with hybrid approaches in cloud (2021)
  12. LogAnomaly: Unsupervised detection of sequential and quantitative anomalies in unstructured logs. (2019)
  13. Stock market prediction using artificial intelligence: A systematic review of systematic reviews (2024)
  14. Self-supervised anomaly detection from distributed traces (2020)
  15. Leveraging Log Instructions in Log-based Anomaly Detection (2022)
  16. Utilizing persistence for post facto suppression of invalid anomalies using system logs (2022)
  17. Stock Price Prediction Using the ARIMA Model (2014) — link
  18. Stock market predication using a linear regression (2017) — link
  19. Survey of State-of-the-art Log-based Failure Diagnosis (2020) — link
  20. Anomaly detection from system tracing data using multimodal deep learning (2019)
  21. A Comprehensive Survey of Logging in Software: From Logging Statements Automation to Log Mining and Analysis (2021) — link
  22. Detecting Log Anomalies with Multi-Head Attention (LAMA) (2021) — link
  23. Deep learning based log anomaly detection with time differences (2020)
  24. Machine learning based anomaly detection of log files using ensemble learning and self-attention (2021)
  25. Translog: A unified transformer-based framework for log anomaly detection (2021) — link
  26. Quality monitoring and assessment of deployed deep learning models for network AIOps (2021)
  27. PULL: Reactive Log Anomaly Detection Based On Iterative PU Learning (2023) — link
  28. AI for IT Operations (AIOps) on Cloud Platforms: Reviews, Opportunities and Challenges (2023) — link
  29. Deep learning for anomaly detection in log data: A survey (2023)
  30. Aiops: real-world challenges and research innovations (2019)
  31. Robust peak detection algorithm (using z-scores) (2014)
  32. The forthcoming Artificial Intelligence (AI) revolution: Its impact on society and firms (2017)
  33. Sock Shop: A microservice demo application — [link](HTTPS: //github.com/microservices-demo/microservices-demo)
  34. Drain3: An online log template miner — [link](https://github.com/IBM/ Drain3)
  35. BM Watson AIOPs — [link](www.ibm.com/cloud/cloud-peak-for-Watson- AIOps)
  36. Long Short-Term Memory (1997) — link
  37. Learning phrase representations using RNN encoder-decoder for statistical machine translation (2014) — link
  38. Attention is all you need (2017)
  39. What is LSTM - long short term memory? (2024) — link
  40. Convolutional gated recurrent unit–recurrent neural network for state-of-charge estimation of lithium-ion batteries (2019)

0 of 48 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.