Jul 2026· 2026 International Conference on Electronics, Computing, Communication and Control Technology (ICECCC)· pp. 1-6· 0 citations· 20 references
Abstract
This study compares Support Vector Regression (SVR) and Long Short-Term Memory (LSTM) models for forecasting annual West Texas Intermediate (WTI) crude oil prices using data from 1981-2023, with projections to 2035. The US Dollar Index (DXY) is incorporated as an explanatory variable to capture exchange-rate effects in global oil markets. A walk-forward crossvalidation framework is employed, and forecasting performance is evaluated using MSE, RMSE, MAE, MAPE, and $\mathrm{R}^{2}$. Results reveal a moderate negative correlation between WTI prices and the DXY index. Forecast comparison tests, including the paired t-test, Wilcoxon signed-rank test, and Diebold-Mariano (DM) test, consistently show that SVR outperforms LSTM. Incorporating DXY further improves forecasting accuracy, particularly for SVR. The extended SVR model achieves the highest explanatory power $\left(\mathrm{R}^{2}=0.928\right)$, compared with the baseline SVR $\left(\mathrm{R}^{2}=0.912\right)$, baseline LSTM $\left(\mathrm{R}^{2}=0.726\right)$, and extended LSTM $\left(\mathrm{R}^{2}=0.781\right)$. These findings suggest that SVR augmented with macro-financial information provides a more suitable framework for medium-term energy and fiscal policy analysis.
The findings demonstrate that rigorous leakage-free validation is essential for reliable forecasting research and that, for monthly Robusta coffee prices, increased model complexity does not necessarily yield superior predictive performance.
Dler H Kadir, D. Khalil, Azhin M. Khudhur· Forecasting· 0 citations
The present paper conducts an analysis of how different AI models perform in predicting the daily closing values of the National Stock Exchange (NSE) Indian Large Cap Stocks Index. The paper evaluates seven models, namely Autoregressive Integrated Moving Average (ARIMA), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Random Forest, Extreme Gradient Boosting (XGBoost), Support Vector Regression (SVR), and Prophet, based on 35-month historical Indian Large Cap Stocks data acquired from the NSE website. A strong evaluation method (walk-forward validation) is used to test the forecasting performance. The model was a deep learning model implemented using TensorFlow, and the other models were implemented using scikit-learn. Correlation (rho), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R²) were used to evaluate performance. The results of the Diebold-Mariano tests proved that Long Short-Term Memory (LSTM) predictions were more accurate compared to the individual forecasts of the ARIMA, GRU, and XGBoost models with a standard level of significance. The recurrent models and tree-based learners in a stacked ensemble with a linear meta-learner performed better than all individual models, reducing RMSE by 18.5% and MAE by 20.6% compared with the best single model. These changes happen because the ensemble combines the strengths of different modeling methods. It uses the ability of networks to remember what happened in the past and the ability of tree-based methods to make decisions in a more complex way. This results in accurate predictions when things are changing quickly. For people who trade and analyze the market, this shows that hybrid AI systems can be really useful for predicting what will happen in the future. If the systems are designed carefully, they can help in making decisions about managing risk and allocating resources. Besides, this work suggests the future incorporation of Transformer-based models to increase predictive power, like Informer or Crossformer.
M. Shindhe, Prayag Gokhale· Cureus Journal of Business a...· 0 citations
Stock price forecasting is a key part of investment decision-making, especially for high-capitalization stocks such as PT Bank Central Asia Tbk. (BBCA). Accurate prediction remains challenging because complex market dynamics and nonlinear price movements influence stock prices. This study proposes a multivariate Long Short-Term Memory (LSTM) model that integrates the closing price and the RSI-14 technical indicator as input features to improve accuracy over conventional univariate approaches. We obtained historical data from the Yahoo Finance API covering January 2015 to December 2025, totaling 2,698 trading days after RSI-14 feature engineering. The dataset was split to 80:20 for training and testing, with MinMaxScaler normalization applied only to the training data to prevent data leakage. The LSTM model used 100 neurons with an input shape of (10, 2) and was trained using the Adam optimizer with early stopping at epoch 24 to avoid overfitting. Evaluation results show an RMSE of 169.72 IDR, an MAE of 134.68 IDR, a MAPE of 1.58%, and an R² of 0.9368, indicating a good regression-level fit. Adding the RSI-14 feature improved the regression metrics relative to a univariate closing-price-only LSTM. Given the modest trend-classification accuracy and the lack of trading back testing, transaction costs, or risk-adjusted performance measures, view the results as a methodological contribution rather than direct evidence of practical investment value. Future work could extend this research by exploring hybrid architectures that combine LSTM with attention mechanisms or Transformer-based models, and by validating the approach on other high-capitalization stocks to assess generalizability across market conditions.
Muhammad Jahron, J. A. Widians, Andi Tejawati· TEPIAN· 0 citations
The USD/IDR exchange rate is a key daily barometer of Indonesia's economic health. Accurate forecasting is vital for trade, inflation, and monetary stability. However, its volatile and nonlinear dynamics pose challenges. While research has applied statistical models, machine learning, and deep learning, few studies offer a comprehensive comparison integrating predictive accuracy with model interpretability, particularly using banking stock prices as exogenous predictors. This study addresses this gap by developing and evaluating eight forecasting approaches—a naive random-walk baseline, two statistical models (ARIMA, SARIMAX), three tree-based ensembles (Random Forest, XGBoost, LightGBM), and two recurrent neural networks (LSTM, GRU) using daily data from January 2015 to July 2026 (3,005 observations). Five major banking stocks (BBCA, BBRI, BMRI, BBNI, BDMN) are included as one-day-lagged exogenous features. Models are assessed via hold-out testing and five-fold walk-forward cross-validation using RMSE, MAE, MAPE, and R². Contrary to expectations, the naive random-walk consistently achieves the lowest error (RMSE=115.24, MAE=63.80, MAPE=0.39%) and the most stable performance, with LSTM as the best-performing complex model (RMSE=269.59, R²=0.785). Diebold-Mariano tests confirm statistical significance (p<0.001). To enhance transparency, SHAP-based Explainable AI is applied to Random Forest, revealing that the lagged USD/IDR value overwhelmingly dominates predictions (mean |SHAP|=1,509.01), while banking stock contributions are negligible. These findings also empirically confirm the well-known Meese-Rogoff puzzle and weak-form market efficiency for USD/IDR, clearly proving that simple baselines remain formidable benchmarks for short-horizon forecasts. This study ultimately underscores the critical importance of combining rigorous benchmarking with XAI to deliver accurate and interpretable predictions for economic policymakers and financial practitioners.
D. Setyawan, Astrid Sulistya Azahra, Mugi Lestari· International Journal of Mat...· 0 citations
Inflation forecasting remains a complex problem due to nonlinear dynamics and interactions among macroeconomic variables, particularly in emerging economies such as Indonesia. Previous studies using deep learning models, including Long Short-Term Memory (LSTM), have shown promising results but often suffer from high variance and limited robustness, especially when temporal dependencies are not properly preserved. This study aims to develop a more stable and accurate forecasting model by integrating Bagging with a Stacked LSTM architecture using the Moving Block Bootstrap (MBB) method. The proposed model utilizes multivariate time series data consisting of inflation, exchange rate (USD/IDR), BI interest rate, and money supply, with preprocessing techniques including Z-score normalization and sliding window transformation. Experimental results show that the model achieves an RMSE of 0.4273 and MAE of 0.3048, indicating good predictive performance. Compared to baseline models such as ARIMA and single LSTM, the proposed approach provides more stable and consistent forecasting results. The model is also able to generate reliable predictions for the next 12 periods, demonstrating its ability to capture temporal patterns effectively. These findings suggest that the integration of Bagging, Stacked LSTM, and MBB improves model robustness and forecasting accuracy. The proposed approach can support data-driven decision-making in economic policy, although further research is needed to incorporate additional variables and explore more advanced forecasting architectures.
This study evaluates whether historical daily trading data combined with simple moving average (SMA) features can improve the accuracy of Long Short-Term Memory (LSTM) neural networks in forecasting short-term stock prices, and whether forecasting accuracy differs systematically between the technology and pharmaceutical sectors. Daily open, high, low, close, and volume data were retrieved through the yfinance Python library for six publicly traded firms: three technology companies (Apple, Alphabet, and NVIDIA) and three pharmaceutical companies (AbbVie, Eli Lilly, and Pfizer). Simple moving averages over 10-, 20-, and 50-day windows were computed from closing prices and supplied to the models as additional inputs. LSTM architectures were implemented in both TensorFlow and PyTorch, trained on chronologically ordered data using an 80/10/10 train-validation-test split, and evaluated with mean squared error, root mean squared error, and mean absolute error. Ten-day out-of-sample forecasts were generated for closing price, daily high, daily low, and trading volume beginning July 21, 2025, and were compared against realized market values. Across the evaluation window, the pharmaceutical sector produced a lower mean absolute percentage error (3.06 percent) than the technology sector (7.27 percent), with AbbVie the most accurately forecast security (2.01 percent) and NVIDIA the least (12.33 percent). Directional accuracy remained near or below chance for every security, indicating that low percentage error reflects proximity to a slow-moving price level rather than a reliable ability to anticipate day-to-day movement. The results indicate that moving averages help LSTM models track the general level and direction of price series in stable, low-volatility conditions, but that models trained exclusively on historical price information cannot account for the exogenous news, regulatory, and innovation shocks that drive a substantial share of price variance, particularly in the technology sector.
Manav Patel· International Journal For Mu...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
