Short-term Memory Model Research Articles

Performance enhancement of mid-infrared NH3 sensor using 9.06 μm QCL based on spectral optimization and NGO-LSTM model.

A detection sensor for mid-infrared ammonia (NH3) has been developed according to wavelength modulation spectroscopy-tunable diode laser absorption spectroscopy (WMS-TDLAS) technology, which can be applied in the chemical and aquaculture industries. A9.06 µm quantum cascade laser (QCL) and a41.5 m multipass gas cell (MPGC) were used to increase the detection limit of NH3. Spectral optimization and theNGO-LSTM (northern goshawk optimization-long short-term memory) model applied to gas detection are designed to improve the accuracy of sensor. Among them, the design of the temperature compensation and spectral drift correction reduces the effect of temperature and other environmental factors. Theoriginal second harmonic signal was denoised using the CEEMDAN-WPD (complete ensemble empirical mode decomposition with adaptive noise-wavelet packet decomposition) algorithm. And the NGO-LSTM algorithm was appliedto NH3 concentration inversion, adaptively optimizing theweight parameters. The experiment reflects that the measured value of the sensor has an excellent linear relationship with the set value (R2 0.9992). The long-term stability of the sensor was verifiedbased on 400 ppb NH3, with an RMSE (root mean square error) of 4.754 ppb. Allan-Werle bias analysis shows that the detection limit (LoD) is approximately 792 ppt at an integration time of 232 s. Subsequent response time and atmospheric environment simulation experiments have proven that this sensor provides an efficient approach for real-time monitoring of NH3.

Functional Brain Network Disruptions in Parkinson's Disease: Insights from Information Theory and Machine Learning.

Objectives: This study investigates disruptions in functional brain networks in Parkinson's Disease (PD), using advanced modeling and machine learning. Functional networks were constructed using the Nonlinear Autoregressive Distributed Lag (NARDL) model, which captures nonlinear and asymmetric dependencies between regions of interest (ROIs). Key network metrics and information-theoretic measures were extracted to classify PD patients and healthy controls (HC), using deep learning models, with explainability methods employed to identify influential features. Methods: Resting-state fMRI data from the Parkinson's Progression Markers Initiative (PPMI) dataset were used to construct NARDL-based networks. Metrics, such as Degree, Closeness, Betweenness, and Eigenvector Centrality, along with Network Entropy and Complexity, were analyzed. Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Long Short-Term Memory (LSTM) models, classified PD and HC groups. Explainability techniques, including SHAP and LIME, identified significant features driving the classifications. Results: PD patients showed reduced Closeness (22%) and Betweenness Centrality (18%). CNN achieved 91% accuracy, with Network Entropy and Eigenvector Centrality identified as key features. Increased Network Entropy indicated heightened randomness in PD brain networks. Conclusions: NARDL-based analysis with interpretable deep learning effectively distinguishes PD from HC, offering insights into neural disruptions and potential personalized treatments for PD.

Vision-Based Prediction of Flashover Using Transformers and Convolutional Long Short-Term Memory Model

The prediction of fire growth is crucial for effective firefighting and rescue operations. Recent advancements in vision-based techniques using RGB vision and infrared (IR) thermal imaging data, coupled with artificial intelligence and deep learning techniques, have shown promising solutions to be applied in the detection of fire and the prediction of its behavior. This study introduces the use of Convolutional Long Short-term Memory (ConvLSTM) network models for predicting room fire growth by analyzing spatiotemporal IR thermal imaging data acquired from full-scale room fire tests. Our findings revealed that SwinLSTM, an enhanced version of ConvLSTM combined with transformers (a deep learning architecture based on a new mechanism called multi-head attention) for computer vision purposes, can be used for the prediction of room fire flashover occurrence. Notably, transformer-based ConvLSTM deep learning models, such as SwinLSTM, demonstrate superior prediction capability, which suggests a new vision-based smart solution for future fire growth prediction tasks. The main focus of this work is to perform a feasibility study on the use of a pure vision-based deep learning model for analysis of future video data to anticipate behavior of fire growth in room fire incidents.

Evaluating the effectiveness of self-attention mechanism in tuberculosis time series forecasting

BackgroundWith the increasing impact of tuberculosis on public health, accurately predicting future tuberculosis cases is crucial for optimizing of health resources and medical service allocation. This study applies a self-attention mechanism to predict the number of tuberculosis cases, aiming to evaluate its effectiveness in forecasting.MethodsMonthly tuberculosis case data from Changde City between 2010 and 2021 were used to construct a self-attention model, a long short-term memory (LSTM) model, and an autoregressive integrated moving average (ARIMA) model. The performance of these models was evaluated using three metrics: root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE).ResultsThe self-attention model outperformed the other models in terms of prediction accuracy. On the test set, the RMSE of the self-attention model was approximately 7.41% lower than that of the LSTM model, MAE was reduced by about 10.99%, and MAPE was reduced by approximately 9.87%. Compared to the ARIMA model, RMSE was reduced by about 28.86%, MAE by about 32.22%, and MAPE by approximately 29.89%.ConclusionThe self-attention model can effectively improve the prediction accuracy of tuberculosis cases, providing guidance for health departments optimizing of health resources and medical service allocation.

Integrating Convolutional Attention and Encoder–Decoder Long Short-Term Memory for Enhanced Soil Moisture Prediction

Soil moisture is recognized as a crucial variable in land–atmosphere interactions. This study introduces the Convolutional Attention Encoder–Decoder Long Short-Term Memory (CAEDLSTM) model to address the uncertainties and limitations inherent in traditional soil moisture prediction methods, especially in capturing complex temporal dynamics across diverse environmental conditions. Unlike existing approaches, this model integrates convolutional layers, an encoder–decoder framework, and multi-head attention mechanisms for the first time in soil moisture prediction. The convolutional layers capture local spatial features, while the encoder–decoder architecture effectively manages temporal dependencies. Additionally, the multi-head attention mechanism enhances the model’s ability to simultaneously focus on multiple key influencing factors, ensuring a comprehensive understanding of complex environmental variables. This synergistic combination significantly improves predictive performance, particularly in challenging climatic conditions. The model was validated using the LandBench1.0 dataset, which includes multiple high-resolution datasets, such as ERA5-land, ERA5 atmospheric variables, and SoilGrids, covering various climatic regions, including high latitudes, temperate zones, and tropical areas. The superior performance of the CAEDLSTM model is evidenced by comparisons with advanced models such as AEDLSTM, CNNLSTM, EDLSTM, and AttLSTM. Relative to the traditional LSTM model, CAEDLSTM achieved an average increase of 5.01% in R2, a 12.89% reduction in RMSE, a 16.67% decrease in bias, and a 4.35% increase in KGE. Moreover, it effectively addresses the limitations of traditional deep learning methods in challenging climates, including tropical Africa, the Tibetan Plateau, and Southeast Asia, resulting in significant enhancements in predictive accuracy within these regions, with R2 values improving by as much as 20%. These results underscore the capabilities of CAEDLSTM in capturing complex soil moisture dynamics, demonstrating its considerable potential for applications in agriculture and water resource monitoring across diverse climates.

Enhancing wave energy farm efficiency: Eigen-stacking ensemble framework

The significant wave height (SWH) plays a pivotal role across diverse domains, ranging from the assessment of wave energy potential to the optimization of marine operations and the advancement of ocean engineering techniques. Understanding the spatial and temporal distribution patterns of SWH data among buoy network stations holds utmost importance for various applications. This research introduces a novel methodology for accurate spatiotemporal SWH prediction across multiple stations within buoy monitoring networks. Leveraging eigen SWH time series enables the identification of dataset patterns, feature extraction, and dimensionality reduction. Examining eigen time series provides insights into SWH dynamics, allowing exclusive use for prediction across all stations. By incorporating this eigen SWH time series as the sole input into a stacking ensemble model, we introduce a highly efficient and accurate framework for SWH prediction. Notably, the methodology achieves success in forecasting hourly SWH values along the western United States coastline, utilizing a stacking ensemble technique for enhanced accuracy. Evaluation metrics confirm outstanding performance, with CE and R2 values surpassing 0.95 and 0.93, respectively. Further, a single Long Short-Term Memory (LSTM) model is incorporated as a benchmark to assess the effectiveness of the stacking ensemble model for significant wave height (SWH) prediction, with results demonstrating that the ensemble model outperforms the single LSTM in terms of accuracy and robustness. Consequently, the introduced stacking ensemble model promises autonomy in forecasting spatial-temporal SWH patterns, extending its applicability within ocean engineering and scientific research.

An optimized bidirectional vision transformer based colorectal cancer detection using histopathological images

The high death rate of colorectal cancer has a long-term impact on human life around the world. Early detection of colorectal cancer leads to an increase in the survival rate of the patients. Overfitting issues occurred due to the presence of imbalanced datasets. To resolve the challenges, a novel deep learning-based approach based on an effective optimization strategy with a bidirectional vision transformer model for colorectal cancer detection. Disease classification is performed using two datasets: colorectal histology images and the NCT-CRC-HE-100 K dataset. Initially, the histopathological images from the dataset are trained and pre-processed using a Trimmed Pixel density-based median filter (TPDMF), which removes noise from the input images and enhances the quality of the images. The features of the pre-processed image are extracted using the Capsule Assisted Res2Net (CAR2N) model, and an optimized Bidirectional Vision Transformer (OBVT) model is presented to identify colorectal cancer. Here, the Bi- Long short-term memory (Bi-LSTM) model is used in the Multi-layer perception module of the vision transformer to reduce the complexity of the detection process. The proposed methodology uses a chaotic sequence-based Snake optimization algorithm (Ch-SOA) to tune the hyperparameters in the proposed model. The proposed model can be analyzed using different performance metrics like accuracy, precision, sensitivity, specificity, False negative rate (FNR), and False positive rate (FPR). The proposed model can obtain an accuracy of 97.11458 % for the colorectal histology images dataset and 97.01235 % for the NCT-CRC-HE-100 K dataset. From this analysis, the proposed model can obtain better results than other existing models.

Rainfall Modeling using a Machine Learning Approach as Support for the Early Warning System for Flood Disasters in Malang Regency, Indonesia

Malang is one of the areas in East Java - Indonesia that has the potential to experience flooding caused by high rainfall. Floods have a negative impact on both agriculture and public health. Therefore, it is necessary to create a rainfall prediction model in Malang that supports the creation of an early warning system for flood disasters. In this research, a machine learning approach was used, namely the Long Short Term Memory model, to model monthly rainfall at three stations in Malang Regency, namely Abd. Saleh Airbase Station, Karangkates Station and Karangploso Station. Several inputs were tried, namely rainfall lag from the station itself and from neighboring stations as well as other weather variables, namely air humidity, air pressure and air temperature. The results obtained from this research are the best LSTM model for Abd. Saleh Airbase Station rainfall in is similar to the best LSTM model for Karangploso Station rainfall, namely a LSTM model with input rainfall lag 6 for the station itself. Meanwhile, the best model for Karangkates Station rainfall is a model with input rainfall lag 1, lag 4 and lag 6 from the three stations.. KEYWORDS :Machine learning, Long short term memory, Monthly rainfall, Modeling.

Comparative analysis of artificial intelligence models for real-time and future forecasting of environmental conditions: A wood-frame historic building case study

Precise environmental monitoring within historic buildings is crucial for ensuring optimal conditions that guarantee the proper preservation of these structures, thereby maintaining their integrity and cultural legacy. Moreover, outdoor environmental conditions can potentially impact the indoor environment, especially in historic structures with deficient insulation materials and damaged envelopes. As climate change points to more recurrent and severe extreme climatic events in the coming years, a current and future comprehensive evaluation of indoor microclimate in historic structures is essential. This study investigated the utilization of machine learning and deep learning algorithms for real-time and future forecasting of the indoor microclimate, including air temperature, relative humidity, and dew point, in a historic building located in San Antonio, Texas, USA. In situ monitored data from April 2022 to January 2023 were used to train different predictive algorithms. The results indicated that Multi-Layer Perceptrons and Support Vector Machine models yielded the most accurate values in terms of real-time forecasting of the indoor microclimate, while Extreme Gradient Boosting model excelled in convergence time. Additionally, Long Short-Term Memory models were the most accurate in predicting indoor microclimate using future weather data for the current century, specifically for 2050 and 2080. The methodology developed in this study can be applied to different construction types and locations globally. As it enables the prediction of environmental conditions crucial for historic preservation, the results have the potential to assist experts in making informed decisions about conserving historic structures, both now and in the future.

Regarding reservoirs as switches: Deriving restored and regulated runoff with one long short-term memory model

Study regionUpper and middle Yangtze River, China Study focusSince measured runoff is no longer natural due to reservoirs, runoff series should be restored (without reservoirs) and regulated (with current reservoirs) for consistency. Restored and regulated runoff from one model benefits from continuously measured data and consistent model conditions. However, current restoration and regulation use two separate models, with data split by reservoir operations. To address this issue, a Reservoir-State Long Short-Term Memory (RS-LSTM) model is proposed to restore and regulate runoff simultaneously by regarding reservoirs as switches. Three schemes are compared: (1) binary states, simulating runoff either with or without reservoirs; (2) dynamic states, beginning from zero and increasing one at each time step after initial impoundment to represent time-dependent operation; (3) no states, namely an LSTM model, as a benchmark. New hydrological insights for the region(1) Nash-Sutcliffe efficiencies over test periods are 0.79, 0.67, and 0.75 for the models using binary, dynamic, and no states, respectively. (2) The binary RS-LSTM model can restore the runoff without historical operation data, validated with the water balance model. (3) The RS-LSTM model is used to evaluate the reservoirs’ impact on downstream lakes through restored and regulated runoff. The comparison indicates that reservoirs alleviate downstream droughts instead of causing low water levels. The proposed RS-LSTM model effectively evaluates the reservoirs’ impacts on downstream areas.

A novel surface deformation prediction method based on AWC-LSTM model

Severe surface deformation can damage the ecological environment, trigger geological disasters, and threaten human life and property. Reliable surface deformation prediction is conducive to reducing potential risks and mitigating disaster losses. Currently, machine learning-based surface deformation prediction models have shown significant improvements in prediction performance. However, most prediction models do not sufficiently consider the characteristics of surface deformation, exhibit subjectivity in parameter settings, and inadequately capture local features in time series data. We introduce the AWC-LSTM model to predict surface deformation. Initially, leveraging the strengths of the autoregressive integrated moving average (ARIMA) model in handling linear signals, the obtained surface deformation information is decomposed to linear and nonlinear parts, and the linear part is predicted. Secondly, by incorporating convolutional neural network (CNN) layers into the long short term memory (LSTM) model, the ability to learn local features is enhanced and the whale optimization algorithm (WOA) is introduced to determine the optimal hyperparameters of the model, thereby predicting nonlinear deformation. The proposed AWC-LSTM model was validated using the Shilawusu coal mine and Beijing as case studies. The outcomes indicate that the deformation predictions for the Shilawusu coal mine and Beijing exhibit a high degree of consistency with the monitored data, with root mean square errors (RMSE) not exceeding 3 mm. This underscores the model’s reliability and applicability across different areas. Comparisons with existing prediction models indicate that the AWC-LSTM model achieves higher predictive accuracy, with an average improvement in accuracy ranging from 28.38 % to 80.59 % over other models.

Detecting Ineffective Efforts during Expiration for Neonates with Attention RNNs

Abstract Patient-ventilator asynchronies occur during mechanical ventilation when there is a mismatch between the patient’s needs and the ventilator’s settings. Ineffective Efforts during Expiration (IEE) is such an asynchrony, which if undetected can cause patient stress and prolong the ventilation duration. Current approaches to detect IEEs only target adult patients and are not directly applicable to newborns. This work explores the use of Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) models to detect IEEs in neonates. An attention mechanism is employed to additionally offer a visual explanation for the network’s classification. All of the tested attention RNN architectures yield an accuracy of over 90%, with LSTMs performing slightly better than simple RNNs. A user study is conducted to evaluate the usability of the employed explainability method. The results show that the used visualizations are intuitive to understand, however the network’s attention itself can be misleading in certain cases.

Efficient cross-lingual plagiarism detection using bidirectional and auto-regressive transformers

<span lang="EN-US">The pervasive availability of vast online information has fundamentally altered our approach to acquiring knowledge. Nevertheless, this wealth of data has also presented significant challenges to academic integrity, notably in the realm of cross-lingual plagiarism. This type of plagiarism involves the unauthorized copying, translation, ideas, or works from one language into others without proper citation. This research introduces a methodology for identifying multilingual plagiarism, utilizing a pre-trained multilingual bidirectional and auto-regressive transformers (mBART) model for document feature extraction. Additionally, a siamese long short-term memory (SLSTM) model is employed for classifying pairs of documents as either "plagiarized" or "non-plagiarized". Our approach exhibits notable performance across various languages, including English (En), Spanish (Es), German (De), and French (Fr). Notably, experiments focusing on the En-Fr language pair yielded exceptional results, with an accuracy of 98.83%, precision of 98.42%, recall of 99.32%, and F-score of 98.87%. For En-Es, the model achieved an accuracy of 97.94%, precision of 98.57%, recall of 97.47%, and an F-score of 98.01%. In the case of En-De, the model demonstrated an accuracy of 95.59%, precision of 95.21%, recall of 96.85%, and F-score of 96.02%. These outcomes underscore the effectiveness of combining the MBART transformer and SLSTM models for cross-lingual plagiarism detection.</span>

A Hybrid Data Preprocessing-Based Hierarchical Attention BiLSTM Network for Remaining Useful Life Prediction of Spacecraft Lithium-Ion Batteries.

As a crucial energy storage for the spacecraft power system, lithium-ion batteries degradation mechanisms are complex and involved with external environmental perturbations. Hence, effective remaining useful life (RUL) prediction and model reliability assessment confronts considerable obstacles. This article develops a new RUL prediction method for spacecraft lithium-ion batteries, where a hybrid data preprocessing-based deep learning model is proposed. First, to improve the correlation between battery capacity and features, the empirically selected high-dimensional features are linearized by using the Box-Cox transformation and then denoised via the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) method. Second, the principal component analysis (PCA) algorithm is employed to perform feature dimensionality reduction, and the output of PCA is further processed by the sliding window technique. Third, a multiscale hierarchical attention bi-directional long short-term memory (MHA-BiLSTM) model is constructed to estimate the capacity in future cycles. Specifically, the MHA-BiLSTM model can predict the RUL of lithium-ion batteries by considering the correlation and significance of each cycle's information during the degradation process on different scales. Finally, the proposed method is validated based on multiple types of experiments under two lithium-ion battery datasets, demonstrating its superior performance in terms of feature extraction and multidimensional time series prediction.

Dynamic long short-term memory model for enhanced product recommendations in e-commerce

Recommendation systems are pivotal for personalized user experiences, employing algorithms to predict and suggest items aligned with user preferences. Deep learning (DL) models, such as recurrent neural networks (RNNs) and long short-term memory networks (LSTMs), excel in capturing sequential dependencies, enhancing recommendation accuracy. However, challenges persist in session-based recommendation systems, particularly with gradient descent and class imbalances. Addressing these challenges, this work introduces dynamic LSTM (D-LSTM), a novel DL-based recommendation system tailored for dynamic E-commerce environments. The primary objective is to optimize recommendation accuracy by effectively capturing temporal dependencies within user sessions. The methodology involves the integration of D-LSTM with weight matrix optimization and a Bayesian personalized ranking (BPR) adaptable learning rate optimizer to enhance learning efficiency. Experimental results demonstrate the efficacy of D-LSTM, showing significant improvements over existing models. Specifically, comparisons with the hybrid time-centric prediction (HTCP) model reveal a performance enhancement of 19.4%, 17.2%, 35.41%, and 21.99% for hit-rate (HR) and mean reciprocal rank (MRR) in 10k and 20k recommendation sets using the Tmall dataset. These findings underscore the superior performance of D-LSTM, highlighting its potential to advance personalized recommendations in dynamic E-commerce settings.

The impacts of the COVID-19 pandemic on the burden of maternal and neonatal disorders: A counterfactual modeling based on the global burden of disease study (2021).

During the COVID-19 pandemic, global health systems faced unprecedented challenges, as well as in maternal and neonatal health, thus this study aims to clarify the impacts of COVID-19 on maternal and neonatal disorders (MNDs), regional variations, and the role of economic support. We have developed a counterfactual model integrating Autoregressive Integrated Moving Average and Long Short-Term Memory models to forecast the burden of MNDs from 2020 To et al., 2021, which was compared with the actual burden to quantify the specific impact of the COVID-19 pandemic on MNDs. During the COVID-19 pandemic, the burden of MNDs surpassed predictions, particularly in Russia, where incidence was about 10.20% higher than expected. In Tokelau, neonatal disorders increased by 412.35% in DALYs. The incidence of maternal disorders in Russia has increased by 12.00%, with maternal abortion and miscarriage increasing by 23.08%. The incidence and prevalence of maternal hypertensive disorders, the incidence of hemolytic disease and other neonatal jaundice and neonatal preterm birth accelerated. In low and low-middle Socio-demographic Index countries, mortality rates from maternal abortion and miscarriage, maternal obstructed labor and uterine rupture, neonatal encephalopathy due to birth asphyxia and trauma significantly increased. Similarly, countries with a low economic support index saw higher burden for these conditions, with the burden decreasing as economic support improved. The COVID-19 pandemic has disproportionately increased the burden of MNDs in countries with lower economic support, highlighting the critical need for strengthened global economic support, particularly in low- and middle-income countries.

Trajectory and Motion Prediction of Autonomous Vehicles Driving Assistant System using Distributed Discriminator Based Bi-Directional Long Short Term Memory Model

Trajectory and Motion Prediction of Autonomous Vehicles Driving Assistant System using Distributed Discriminator Based Bi-Directional Long Short Term Memory Model

Prediction of ROI Achievements and Potential Maximum Profit on Spot Bitcoin Rupiah Trading Using K-means Clustering and Patterned Dataset Model

Since Satoshi Nakamoto first proposed the idea of bitcoin in 2009, the cryptocurrency and prediction methods for it have grown and changed exceptionally quickly. The Patterned Dataset Model was a valuable tool in earlier studies to explain how changes in the price of Bitcoin affect the movements of other cryptocurrencies in a digital trading market. Three different kinds of datasets are generated by this model: patterned datasets under full conditions, patterned datasets under dropping prices (Crash), and patterned datasets under rising prices (Moon). The K-means approach was then used to cluster these three datasets. Specifically, each dataset was split into two clusters, and the clustering score was determined by utilizing eight unique clustering metrics. Consequently, the best clustering score was found in the patterned dataset in the crash situation. Additionally, from 2022 to 2024, the raw data from this crash-condition-patterned dataset is used to determine the possibility of reaching maximum profit and return on investment (ROI) daily and monthly. According to the calculation results, the range computed over the course of a whole month (30 to 31 days) is significantly larger than the daily range (24 hours multiplied by one month), which represents the most significant profit and ROI attained before the emergence of the first diamond crash level. This research also covers the application of a deep learning model to forecast patterned datasets for crash scenarios that may occur many days in advance. The ConvLSTM2D Model performs better in predicting pattern dataset values for the subsequent crash scenario, according to the hyperparameter comparison between the Gated Recurrent Unit (GRU) Model and the 2D Convolutional Long Short-Term Memory Model.

An Improved Parallel Heterogeneous Long Short-Term Memory Model with Bayesian Optimization for Time Series Prediction

Currently, Deep Learning (DL) with the Recurrent Neural Networks (RNN) variants is being applied successfully in many domains of Engineering for prediction. In view of the demand for precise forecasting and the aid of Artificial Intelligence Tools, time series prediction reveals a vital task in decision-making and risk assessment. However, the application of novel Recurrent DL models for obtaining an accurate prediction of time series is yet to be explored. Recent trends reveal that Hybrid Neural Networks and DL models are appropriate for time series forecasts. At the same time, the model's selection and the hyperparameter's tuning can greatly impact its performance. To address this problem, a parallel long-term memory (PLSTM) model integrated with Bayesian hyperparameter optimization (PLSTM-BO) is proposed for time series prediction. The model is tuned in terms of key parameters, including the number of neurons, dropout, learning rate, and optimization technique. The model's performance is assessed using the SARS-COVID-19 cumulative cases, deaths, recovery cases, and NIFTY 50 stock closing price time series dataset. The obtained results convey that the current model exhibits remarkable performance compared to existing models.

Hybrid CNN-GRU Model for Real-Time Blood Glucose Forecasting: Enhancing IoT-Based Diabetes Management with AI.

For people with diabetes, controlling blood glucose level (BGL) is a significant issue since the disease affects how the body metabolizes food, which makes careful insulin regulation necessary. Patients have to manually check their blood sugar levels, which can be laborious and inaccurate. Many variables affect BGL changes, making accurate prediction challenging. To anticipate BGL many steps ahead, we propose a novel hybrid deep learning model framework based on Gated Recurrent Units (GRUs) and Convolutional Neural Networks (CNNs), which can be integrated into the Internet of Things (IoT)-enabled diabetes management systems, improving prediction accuracy and timeliness by allowing real-time data processing on edge devices. While the GRU layer records temporal relationships and sequence information, the CNN layer analyzes the incoming data to extract significant features. Using a publicly accessible type 1 diabetes dataset, the hybrid model's performance is compared to that of the standalone Long Short-Term Memory (LSTM), CNN, and GRU models. The findings show that the hybrid CNN-GRU model performs better than the single models, indicating its potential to significantly improve real-time BGL forecasting in IoT-based diabetes management systems.