Multi-Source Financial Data Integration via LSTM-GRU-Attention Models for Robust Forex Predictions

Electricity load forecasting is a crucial undertaking within all the deregulated markets globally. Among the research challenges on a global scale, the investigation of deep transfer learning (DTL) in the field of electricity load forecasting represents a fundamental effort that can inform artificial intelligence applications in general. In this paper, a comprehensive study is reported regarding day-ahead electricity load forecasting. For this purpose, three sequence-to-sequence (Seq2seq) deep learning (DL) models are used, namely the multilayer perceptron (MLP), the convolutional neural network (CNN) and the ensemble learning model (ELM), which consists of the weighted combination of the outputs of MLP and CNN models. Also, the study focuses on the development of different forecasting strategies based on DTL, emphasizing the way the datasets are trained and fine-tuned for higher forecasting accuracy. In order to implement the forecasting strategies using deep learning models, load datasets from three Greek islands, Rhodes, Lesvos, and Chios, are used. The main purpose is to apply DTL for day-ahead predictions (1–24 h) for each month of the year for the Chios dataset after training and fine-tuning the models using the datasets of the three islands in various combinations. Four DTL strategies are illustrated. In the first strategy (DTL Case 1), each of the three DL models is trained using only the Lesvos dataset, while fine-tuning is performed on the dataset of Chios island, in order to create day-ahead predictions for the Chios load. In the second strategy (DTL Case 2), data from both Lesvos and Rhodes concurrently are used for the DL model training period, and fine-tuning is performed on the data from Chios. The third DTL strategy (DTL Case 3) involves the training of the DL models using the Lesvos dataset, and the testing period is performed directly on the Chios dataset without fine-tuning. The fourth strategy is a multi-task deep learning (MTDL) approach, which has been extensively studied in recent years. In MTDL, the three DL models are trained simultaneously on all three datasets and the final predictions are made on the unknown part of the dataset of Chios. The results obtained demonstrate that DTL can be applied with high efficiency for day-ahead load forecasting. Specifically, DTL Case 1 and 2 outperformed MTDL in terms of load prediction accuracy. Regarding the DL models, all three exhibit very high prediction accuracy, especially in the two cases with fine-tuning. The ELM excels compared to the single models. More specifically, for conducting day-ahead predictions, it is concluded that the MLP model presents the best monthly forecasts with MAPE values of 6.24% and 6.01% for the first two cases, the CNN model presents the best monthly forecasts with MAPE values of 5.57% and 5.60%, respectively, and the ELM model achieves the best monthly forecasts with MAPE values of 5.29% and 5.31%, respectively, indicating the very high accuracy it can achieve.

The integration of Artificial Intelligence (AI) into financial forecasting has transformed traditional stock market prediction methods. This research paper explores the effectiveness of AI techniques in forecasting stock trends within the Indian stock market over the last few years. We examine AI methodologies, including machine learning (ML), deep learning (DL), and hybrid models applied to the Bombay Stock Exchange (BSE) and the National Stock Exchange (NSE). By comparing historical data with AI-predicted trends, this study evaluates prediction accuracy and market relevance (Patel et al., 2015). Furthermore, the research outlines existing study gaps and proposes a future scope of integrating AI with behavioral finance and real-time analytics (Chen et al., 2022). The prediction of stock market trends remains a significant challenge due to the stochastic and non-linear nature of financial time series data (Zhang & Zhou, 2020). With the proliferation of Artificial Intelligence (AI), particularly Machine Learning (ML) and Deep Learning (DL) techniques, there has been a paradigm shift in the modeling and forecasting of stock price movements (Fischer & Krauss, 2018). This paper presents a comprehensive study on the effectiveness of AI in predicting stock market trends within the Indian financial ecosystem, focusing on a comparative analysis of AI models implemented over the last five years (2018–2023) on major indices and stocks listed on the NSE and BSE. This research evaluates the performance of multiple AI algorithms—including Support Vector Machines (SVM), Random Forests (RF), Long Short-Term Memory (LSTM) networks, Gated Recurrent Units (GRU), and hybrid ensemble models—in forecasting short- and medium-term price trends using historical stock data (Krauss et al., 2017; Chen & He, 2021). Standard evaluation metrics such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and classification accuracy are employed to assess model efficacy. The results indicate that while traditional machine learning algorithms offer moderate predictive power, advanced deep learning and hybrid models significantly outperform them, particularly during periods of high market volatility, such as the COVID-19 pandemic and post-lockdown recovery phases (Weng et al., 2018; Jain & Jain, 2021). The study identifies key research gaps, including limited sectoral diversification in datasets, inadequate integration of sentiment and behavioral data, overfitting issues in complex models, and the lack of real-time prediction systems in the Indian context (Gupta & Pathak, 2022). Furthermore, current models often ignore nonquantitative factors such as investor sentiment, macroeconomic indicators, and global events, which can critically impact prediction accuracy (Nassirtoussi et al., 2014). These limitations suggest the need for a more holistic and interdisciplinary approach to AI-driven financial forecasting. The scope for future research includes the development of real-time, adaptive AI systems using high-frequency trading data, the incorporation of behavioral finance through social media and news analytics, and the exploration of quantum computing-based AI models (Liu et al., 2023). From a practical standpoint, the findings of this study offer valuable insights for institutional investors, financial analysts, regulatory bodies, and developers of AIpowered trading platforms. The study concludes that while AI is not a definitive solution for market prediction, it provides a powerful augmentative tool that, when designed with robustness, transparency, and adaptability, can significantly enhance decision-making in India’s fastevolving financial markets.

This study aims to compare the effectiveness of two deep learning models, namely Convolutional Neural Network (CNN) and combined CNN with Gated Recurrent Unit (CNN-GRU), in predicting Petrobras stock prices. Using historical stock price data from Yahoo Finance for the period 2017-2023, this study evaluates the performance of both models based on several evaluation metrics. The results showed that the CNN-GRU model outperformed the pure CNN model in all evaluation metrics, with an increase in RMSE value of 4.17% and an increase in R² value of 0.47% on the test data. The CNN-GRU model achieved 96.14% accuracy on the test data, while the CNN model achieved 96.04%. These findings indicate that the integration of CNN's feature extraction capabilities with GRU's temporal dependency modeling capabilities can improve stock price prediction accuracy. This research contributes to the computational finance literature by presenting an in-depth comparative analysis of the application of hybrid deep learning architectures in stock market prediction.

Stock prices are a non-linear, long-term series of data. The research topic of this paper is to use the Long Short-Term Memory Network (LSTM) model to analyze the historical stock price data of BYD (BYD) to predict the future stock price trend, and to provide intelligent reference for investors by evaluating the model performance, aiming to improve the accuracy of stock price prediction. The accuracy and consistency of stock price prediction can be increased by using this method, which successfully takes advantage of the time characteristics in historical data. After explaining the idea and architecture of LSTM, this article creates a stock price prediction model based on it, using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) as evaluation metrics. At the same time, network structures with different layers were designed to verify the effectiveness of the model, and the experimental results were shown. The advantages and limitations of the model are highlighted, and the feasibility of deep learning in the stock financial market and the credibility and advantages of strategic decision-making are demonstrated.

This article investigates the identification of hate speech on social media using machine learning and deep learning techniques. The research uses metrics such as F-measure, AUC-ROC, precision, accuracy, and recall assessing the effectiveness of various tactics. The findings indicate that deep learning models, particularly the bidirectional long short-term memory (BiLSTM) architecture, consistently outperform other methods in categorization tasks. The research emphasizes the importance of sophisticated neural network designs in identifying the intricacies of hostile and offensive content online. The study offers insights for promoting early identification and prevention of cyberbullying, improving secure and inclusive online environments. Future research may explore real-time detection systems, hybrid approaches, or the integration of complementary components to enhance and improve innovative technology in tackling this significant social issue. A sample tweet was annotated by specialists who categorize tweets as hate speech, offensive language, or neutral. The researchers applied shallow learning methodologies and integrated word embeddings like Word2Vec and GloVe to enhance the efficacy of deep learning models. The results indicate that BiLSTM surpasses shallow learning methods in detecting hate speech on Twitter, highlighting the efficacy of deep learning approaches in recognizing and tracking hate speech on social media platforms. When comparing different deep learning and machine learning models on different datasets, the results reveal that deep learning techniques are usually more effective. A reasonably high level of accuracy is achieved by KNN and SVM among classical algorithms, whereas Naïve Bayes performs the poorest. While deep learning approaches provide better results, tree-based models such as Random Forest and Decision Trees offer more consistent accuracy. Models based on neural networks, such as LSTM, CNN, and BI-LSTM, perform well, with LSTM-based methods excelling in particular. The most successful strategy for classification problems is the model presented, which obtains the greatest accuracy, precision, recall, F1-score of 95%. The research aids in the development of advanced tools and methodologies to mitigate hate speech on social media and foster positive online interactions. Future research may investigate alternative deep learning architectures, such as transformers, to enhance hate speech detection efficacy. The advancement of interpretable AI methodologies for identifying hate speech and delivering transparent forecasts might enhance user confidence and facilitate better content moderation decisions.

Precision cancer therapy relies on identifying tumor-specific genetic dependencies, which are genes critical for cancer cell survival and proliferation that can be leveraged as therapeutic vulnerabilities. The Cancer Dependency Map (DepMap) datasets from the Broad and Sanger Institutes provide a comprehensive resource for systematically uncovering these dependencies through genome-wide loss-of-function CRISPR screens. These datasets have enabled the development of predictive models for cancer cell lines or tumors that cannot be directly screened. Recent machine and deep learning approaches have utilized gene expression data, with or without the integration of other omics profiles, to predict such dependencies. However, existing methods often fall short in capturing pathway-level interactions among genes and in effectively integrating insights from both Broad and Sanger DepMap datasets due to limitations in model design and training strategies. To address these challenges, we present a novel deep learning model that integrates pathway-level information and both DepMap datasets to predict the dependency of individual genes in cancer cell lines. In our approach, each cancer cell line is represented by activity scores across thousands of biological pathways and functions, while each gene is modeled based on its involvement in these pathways. This design enables the model to comprehensively capture functional interactions and dependencies within the cellular context, rather than treating genes in isolation. To further enhance the representation of genes targeted for knockout, we implemented a contrastive learning framework. This framework was pretrained on the Sanger CRISPR screens to classify whether two genes exhibit similar dependency profiles and was subsequently fine-tuned on the Broad CRISPR screens to predict the dependency score of each gene in a specific cell line. The final deep learning model demonstrated strong predictive performance, achieving a median correlation coefficient of 0.86 between actual and predicted dependency scores across genes in individual cell lines. The model significantly outperformed baseline configurations that excluded pathway integration or contrastive learning. Notably, cell lines from lung, head and neck, esophageal, and gastric cancers exhibited particularly high prediction accuracy, potentially due to shared biological characteristics within these tumor types. Further investigation is warranted to identify factors contributing to this performance. In summary, our novel deep learning model improves cancer dependency prediction by leveraging pathway-level insights and integrating multiple DepMap datasets. Ongoing efforts include further model refinement, expanded validation experiments, and systematic model interpretation to uncover the underlying mechanisms captured by the model. Citation Format: Chien-Hung Shih, Yu-Chiao Chiu. Pathway-informed deep learning enhances cancer dependency prediction using multi-source CRISPR screening data [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3645.

With the increasing integration of artificial intelligence (AI) in several scientific domains, there is a rising demand for advanced AI tools capable of addressing advanced research challenges. A challenge of paramount importance lies in accurately predicting the streamflow within river basins. Effective river flow prediction holds significant relevance, particularly given the substantial societal implications of river usage, encompassing areas such as transportation, agriculture, and power generation. The present study introduces a novel approach to streamflow prediction involving the development of a Deep Learning (DL) model that combines a convolutional neural network with Transfer Learning (TL) techniques to predict streamflow in river systems. With the aim of training the developed DL model, the study employed a time-series dataset containing hydrological data related to two distinct river basins, i.e., Paraíba do Sul, in Brazil, and Zambezi in the state of Mozambique. The developed DL models exhibited the capability to effectively predict the river flow with a one-day horizon, relying on the preceding three or seven days of historical data. To overcome the limited availability of training data and reduce the training time of DL models, TL was leveraged to incorporate two additional distinct time-series datasets, i.e., historical streamflow data from the São Francisco River in Brazil, and climate data from Delhi, India. The application of TL significantly reduced training time, leading only to a minimal decrease in prediction performance. Indeed, in the case DL models were trained on data collected from the Paraíba do Sul River, a substantial reduction in training time was observed - up to 27% - with a modest percentage decrease of 0.31% in test predictive performance (). Similarly, TL induced a significant reduction in training time of up to 48%, while resulting in a modest 2% reduction in test predictive performance () for the Zambezi dataset. The findings underscore the significance of TL as a strategic and viable approach to improve the efficiency of river flow prediction models in the context of basins with limited hydrological data available.

The electricity market is constantly evolving, being driven by factors such as market liberalization, the increasing use of renewable energy sources (RESs), and various economic and political influences. These dynamics make it challenging to predict wholesale electricity prices. Accurate short-term forecasting is crucial to maintaining system balance and addressing anomalies such as negative prices and deviations from predictions. This paper investigates short-term electricity price forecasting using historical time series data and employs advanced deep learning algorithms. First, four deep learning models are implemented and proposed, which are a convolutional neural network (CNN) with an integrated attention mechanism, a hybrid CNN followed by a gated recurrent unit model (CNN-GRU) with an attention mechanism, and two ensemble learning models, which are a soft voting ensemble and a stacking ensemble model. Also, the optimized version of a transformer model, the Multi-Head Attention model, is introduced. Finally, the perceptron model is used as a benchmark for comparison. Our results show excellent prediction accuracy, particularly in the hybrid CNN-GRU model with attention, thereby achieving a mean absolute percentage error (MAPE) of 6.333%. The soft voting ensemble model and the Multi-Head Attention model also performed well, with MAPEs of 6.125% and 6.889%, respectively. These findings are significant, as previous studies have not shown high performance with transformer models and attention mechanisms. The presented results offer promising insights for future research in this field.

Evaluation of multispectral imaging for freeze damage assessment in strawberries using AI-based computer vision technology

With the presentation of super-resolution convolutional neural network, deep learning approach was applied to image super-resolution reconstruction for the first time. By using convolutional neural network, the deep learning approaches can directly learn the mapping relationship between the low-resolution image and high-resolution image, and have achieved better reconstruction effects than the traditional image super-resolution reconstruction methods. Subsequently, a series of improved deep learning approaches have been proposed, and the reconstruction effects have been improved continuously. This paper systematically summa rizes the image super-resolution reconstruction approaches based on deep learning, analyzes the characteristics of different models, and compares the main deep learning models based on the experiments. Furthermore, based on deep learning model, the future research directions of the image super-resolution reconstruction methods based on deep learning models are reasonably predicted.

PurposeTo conduct a head-to-head comparison between deep learning (DL) and radiomics models across institutions for predicting microvascular invasion (MVI) in hepatocellular carcinoma (HCC) and to investigate the model robustness and generalizability through rigorous internal and external validation.MethodsThis retrospective study included 2304 preoperative images of 576 HCC lesions from two centers, with MVI status determined by postoperative histopathology. We developed DL and radiomics models for predicting the presence of MVI using B-mode ultrasound, contrast-enhanced ultrasound (CEUS) at the arterial, portal, and delayed phases, and a combined modality (B + CEUS). For radiomics, we constructed models with enlarged vs. original regions of interest (ROIs). A cross-validation approach was performed by training models on one center’s dataset and validating the other, and vice versa. This allowed assessment of the validity of different ultrasound modalities and the cross-center robustness of the models. The optimal model combined with alpha-fetoprotein (AFP) was also validated. The head-to-head comparison was based on the area under the receiver operating characteristic curve (AUC).ResultsThirteen DL models and 25 radiomics models using different ultrasound modalities were constructed and compared. B + CEUS was the optimal modality for both DL and radiomics models. The DL model achieved AUCs of 0.802–0.818 internally and 0.667–0.688 externally across the two centers, whereas radiomics achieved AUCs of 0.749–0.869 internally and 0.646–0.697 externally. The radiomics models showed overall improvement with enlarged ROIs (P < 0.05 for both CEUS and B + CEUS modalities). The DL models showed good cross-institutional robustness (P > 0.05 for all modalities, 1.6–2.1% differences in AUC for the optimal modality), whereas the radiomics models had relatively limited robustness across the two centers (12% drop-off in AUC for the optimal modality). Adding AFP improved the DL models (P < 0.05 externally) and well maintained the robustness, but did not benefit the radiomics model (P > 0.05).ConclusionCross-institutional validation indicated that DL demonstrated better robustness than radiomics for preoperative MVI prediction in patients with HCC, representing a promising solution to non-standardized ultrasound examination procedures.

Heart disease, which is one of the most common diseases in the world, is expected to remain the leading cause of mortality on a global scale. Therefore the aim of this study is to classify heart disease using a deep learning approach in an open-access dataset that includes data from patients with and without heart disease. In this study, a deep learning model was applied to an open-access data set containing the data of patients with and without heart disease. The performance of the method used was evaluated with the performance criteria of specificity, sensitivity, accuracy, positive predictive value, and negative predictive value. Specificity, sensitivity, accuracy, positive predictive value and negative predictive value from the performance criteria obtained from the model were calculated as 0.946, 0.903, 0.9245, 0.9436 and 0.907, respectively. As a result of the findings obtained from the study, it was seen that the data set we discussed was successfully classified with the deep learning model used. With this obtained high classification performance, the factors associated with the disease can be revealed.

Objectives: Language detection is the process of identifying a language associated with a text. The proposed system aims to detect the Dravidian language that is associated with the given text using different machine learning and deep learning algorithms. The paper presents an empirical analysis of the results obtained using the different models. It also aims to evaluate the performance of a language agnostic model for the purpose of language detection. Method: An empirical analysis of Dravidian language identification in social media text using machine learning and deep learning approaches with k-fold cross validation has been implemented. The identification of Dravidian languages, including Tamil, Malayalam, Tamil Code Mix, and Malayalam Code Mix, is performed using both machine learning (ML) and deep learning algorithms. The machine learning algorithms used for language detection are Naive Bayes (NB), Multinomial Logistic Regression (MLR), Support Vector Machine (SVM), and Random Forest (RF). The supervised Deep Learning (DL) models used include BERT, mBERT and language agnostic models. Findings: The language agnostic model outperform all other models considering the task of language detection in Dravidian languages. The results of both the ML and DL models are analyzed empirically with performance measures like accuracy, precision, recall, and f1-score. The accuracy associated with different machine learning algorithms varies from 85% to 89%. It is evident from the experimental result that the deep learning model outperformed with an accuracy of 98%. Novelty: The proposed system emphasizes on the use of the language agnostic model to implement the process of detecting Dravidian languages associated with the given text which provides a promising result of 98% accuracy which is higher than the existing methodologies. Keywords: Language, Machine learning, Deep learning, Transformer model, Encoder, Decoder

With the increasing percentage penetration of wind energy, accurate Wind Speed/Power Forecasting (WSF/WPF) has become a crucial tool for power trading and electricity grid operations. The physical and statistical methods used to forecast mostly rely on meteorological information, site specific data and historical data from wind farms. Most of the existing works are limited to utilize single input data source and a prediction variable. However, some gap in using multiple data sources and multiple prediction variables as input has been identified to substantiate and enhance forecast accuracy of these models. In this paper, a multi-source and multivariate surveillance approach has been conducted using time series methods and machine learning algorithms. Firstly, the proposed model uses multiple sources to understand the deciding prediction variables. Secondly, an analogy among time series Autoregressive Integrated Moving Average (ARIMA), Recurrent Neural Networks (RNNs) with Long Short-term Memory units (LSTMs) is used. Although, the prediction models using single data source are economical for forecast providers and wind power producers considering multiple sources can certainly improve the forecasting accuracy thus reducing wind curtailments and imbalance penalties. A comparative approach has been made using data sources obtained for Gulf of Khambhat, a south to north penetration of the Arabian Sea on the western shelf of India. The obtained results show that the forecasting accuracy can be improved by considering multiple sources of input data.

COVID-19 is highly infectious and causes acute respiratory disease. Machine learning (ML) and deep learning (DL) models are vital in detecting disease from computerized chest tomography (CT) scans. The DL models outperformed the ML models. For COVID-19 detection from CT scan images, DL models are used as end-to-end models. Thus, the performance of the model is evaluated for the quality of the extracted feature and classification accuracy. There are four contributions included in this work. First, this research is motivated by studying the quality of the extracted feature from the DL by feeding these extracted to an ML model. In other words, we proposed comparing the end-to-end DL model performance against the approach of using DL for feature extraction and ML for the classification of COVID-19 CT scan images. Second, we proposed studying the effect of fusing extracted features from image descriptors, e.g., Scale-Invariant Feature Transform (SIFT), with extracted features from DL models. Third, we proposed a new Convolutional Neural Network (CNN) to be trained from scratch and then compared to the deep transfer learning on the same classification problem. Finally, we studied the performance gap between classic ML models against ensemble learning models. The proposed framework is evaluated using a CT dataset, where the obtained results are evaluated using five different metrics The obtained results revealed that using the proposed CNN model is better than using the well-known DL model for the purpose of feature extraction. Moreover, using a DL model for feature extraction and an ML model for the classification task achieved better results in comparison to using an end-to-end DL model for detecting COVID-19 CT scan images. Of note, the accuracy rate of the former method improved by using ensemble learning models instead of the classic ML models. The proposed method achieved the best accuracy rate of 99.39%.