Paper: arXiv 2506.09851

Authors: Md. Yeasin Rahat, Rajan Das Gupta, Nur Raisa Rahman, Sudipto Roy Pritom, Samiur Rahman Shakir, Md Imrul Hasan Showmick, Md. Jakir Hossen

Abstract

The prediction of foreign exchange rates, such as the US Dollar (USD) to Bangladeshi Taka (BDT), plays a pivotal role in global financial markets, influencing trade, investments, and economic stability. This study leverages historical USD/BDT exchange rate data from 2018 to 2023, sourced from Yahoo Finance, to develop advanced machine learning models for accurate forecasting. A Long Short-Term Memory (LSTM) neural network is employed, achieving an exceptional accuracy of 99.449%, a Root Mean Square Error (RMSE) of 0.9858, and a test loss of 0.8523, significantly outperforming traditional methods like ARIMA (RMSE 1.342). Additionally, a Gradient Boosting Classifier (GBC) is applied for directional prediction, with backtesting on a $10,000 initial capital revealing a 40.82% profitable trade rate, though resulting in a net loss of $20,653.25 over 49 trades. The study analyzes historical trends, showing a decline in BDT/USD rates from 0.012 to 0.009, and incorporates normalized daily returns to capture volatility. These findings highlight the potential of deep learning in forex forecasting, offering traders and policymakers robust tools to mitigate risks. Future work could integrate sentiment analysis and real-time economic indicators to further enhance model adaptability in volatile markets.

Complexity vs Empirical Score

  • Math Complexity: 6.0/10
  • Empirical Rigor: 7.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced deep learning architectures (LSTM) and specific loss functions (MSE, exponential loss) with some LaTeX notation, indicating moderate mathematical density. It demonstrates high empirical rigor through specific backtesting results on real financial data ($10,000 capital, 49 trades, net loss calculation), use of technical indicators, and comparative performance metrics (RMSE, accuracy) against traditional models.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Forecast USD/BDT Exchange Rates<br>using ML/AI"] --> B{"Data Preparation"}
  
  B --> C["Historical Data<br>2018-2023 (Yahoo Finance)<br>Normalized Daily Returns"]
  
  C --> D{"Model Development"}
  
  D --> E["LSTM Neural Network<br>Time Series Forecasting"]
  D --> F["Gradient Boosting Classifier<br>Directional Prediction"]
  
  E --> G{"Performance Evaluation"}
  F --> G
  
  G --> H["Key Findings & Outcomes"]
  
  H --> I["LSTM Results<br>Accuracy: 99.449%<br>RMSE: 0.9858<br>Outperforms ARIMA (1.342)"]
  H --> J["GBC Backtesting<br>$10,000 Initial Capital<br>40.82% Win Rate<br>Net Loss: -$20,653.25"]
  H --> K["Market Analysis<br>BDT/USD declined 0.012 → 0.009<br>Volatility captured via normalization"]

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 univariate LSTM model for USD/BDT exchange rate forecasting and a Gradient Boosting Classifier for directional prediction.

Source sentence
A Long Short-Term Memory (LSTM) neural network is employed, achieving an exceptional accuracy of 99.449%, a Root Mean Square Error (RMSE) of 0.9858, and a test loss of 0.8523, significantly outperforming traditional methods like ARIMA (RMSE 1.342).

Assumes

No explicit assumptions found in the text.

Data

Backtest sample: USD/BDT · fx · daily

  • USD/BDT exchange rate data · Yahoo Finance market-prices public
    2018 to 2023 · USD/BDT currency pair
    Source sentence
    Yahoo Finance provided the historical USD/BDT exchange rate data from 2018 to 2023, which was then preprocessed with MinMaxScaler and divided into 80% training and 20% testing sets.

Recognised sources: Yahoo Finance

Costs

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

Finds 4

  • The LSTM model achieved 99.449% directional accuracy, RMSE 0.9858 and MAE 0.752, outperforming ARIMA and GBC. performance
    Source sentence
    With an RMSE of 0.986, MAE of 0.752, and Directional Accuracy of 99.449%, the LSTM model fared better than both GBC and ARIMA.
  • The Gradient Boosting Classifier yielded a 40.82% win rate over 49 trades but resulted in a net loss of $20,653.25. negative
    Source sentence
    Backtesting with a $10,000 beginning capital revealed that the GBC, which is utilized for directional prediction, produced a trading performance with a 40.82% win rate over 49 trades
  • ARIMA baseline had RMSE 1.342, MAE 1.015 and directional accuracy 52.30%, trailing behind LSTM and GBC. performance
    Source sentence
    With an RMSE of 1.342, MAE of 1.015, and Directional Accuracy of 52.30%, ARIMA trailed behind the GBC, which had RMSE of 1.128, MAE of 0.894, and Directional Accuracy of 40.00%.
  • Diebold‑Mariano test between LSTM and ARIMA gave p‑value 0.021, indicating statistical significance at the 5% level. statistical
    Source sentence
    For LSTM vs. ARIMA, a Diebold-Mariano (DM) test produced a p-value of 0.021, indicating statistical significance at the 5% level.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 39 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.