Long Short-term Memory Model Research Articles

Using a long short-term memory model to predict force values of Taekwon-do turning based on spatio-temporal parameters.

Purpose: The aim of this study was to investigate the feasibility of using Long Short-Term Memory (LSTM) neural networks to predict Taekwondo kick force from data obtained by inertial measurement unit (IMU) sensors, providing a cost-effective alternative to traditional force plates in sports biomechanics. Methods: IMU (Noraxon Ultium) data from 13 International Taekwon-do Federation (ITF) athletes (9 training, 4 validation) across genders and skill levels (expert in training, expert/advanced in validation) were collected. Sensors were attached to a foot, shank and tight of kicking leg. Athletes performed turning kicks in diverse stances towards a padded force plate (2000 Hz) attached to a wall. LSTM models were trained to predict kick force value, and trained on capturing the IMU data from sensors placed on the lower limb. Results: The trained LSTM models showed accuracy on the training data (R 2 values in the range of 0.972-0.978). Feature validity analysis highlighted the importance of ankle dorsiflexion in shaping the model score. Model performance on the validation dataset was less consistent, ranging from good accuracy (RMSE 6.91) to poor accuracy (RMSE over 30), depending on the participant tested. Conclusions: This study demonstrated the potential of LSTM models combined with IMU data to predict Taekwondo kick forces. Although the validation performance indicated the need for further model refinement or the inclusion of additional input variables, the results highlighted the feasibility of predicting force values without relying on a force plate. This approach could enhance the accessibility of field studies conducted outside laboratory settings.

A CNN-LSTM-attention based seepage pressure prediction method for Earth and rock dams

The objective of seepage pressure monitoring of earth and rock dams is to predict seepage pressure in order to avoid potential risks. However, existing models for predicting seepage pressure in earth and rock dams do not account for the numerous nonlinearities between seepage pressure and the factors that influence it. These models lack the accuracy and generalizability required for effective risk management. In order to address this issue, this paper puts forth a methodology for the prediction of seepage pressure in earth and rock dams. This methodology is based on the use of Convolutional Neural Networks (CNN), Long Short-Term Memory Networks (LSTM), and an attention mechanism. The method initially normalizes each influence factor and divides the dataset. Subsequently, it employs a Convolutional Neural Network (CNN) to extract features from the data. Long Short-Term Memory (LSTM) networks are particularly adept at handling non-smooth time series data, enabling the capture of the deep information embedded within seepage pressure data. Furthermore, the introduction of attention mechanisms allows for the extraction of key information, ultimately enhancing the prediction accuracy and stability. The analysis of engineering examples demonstrates that, in comparison with the single CNN-LSTM, LSTM, Transformer, and BP models, the MAE, MAPE, and RMSE of the proposed method in this paper at two measurement points are the smallest among the four models. The results demonstrate that, in comparison to the other three prediction models, the method exhibits superior prediction accuracy and enhanced stability, is capable of discerning the local variation characteristics of seepage pressure data, exhibits enhanced robustness, and provides a novel approach for accurate prediction and analysis of seepage pressure in earth and rock dams.

Deep learning and similarity-based models for predicting turbofan engine remaining useful life: insights from the CMAPSS dataset

Abstract Turbofan engines are having a growing role in modern aircraft maintenance. Due to this increase, estimation of remaining useful life (RUL) of these engines is an important area of study in the field of reliability and maintenance optimisation. In this work, we propose a hybrid approach that combines deep learning models with similarity-based methods for accurate RUL estimation. For a better comparison, we evaluate four architectures: dropout long short-term memory (LSTM), bidirectional LSTM, convolutional neural network 1D (CNN 1D), and multi-layer LSTM. The FD002 subset of NASA’s Commercial Modular Aero-Propulsion System Simulation dataset is used in the case study. Root mean square error (RMSE) and mean absolute error (MAE) were used for performance metrics. The main output of the study suggests that the dropout LSTM model achieves the best prediction accuracy with an RMSE score of 26.547 and a MAE score of 18.749. It is worth noting that these are achieved despite requiring higher computational resources compared to multi-layer LSTM. Furthermore, all models had difficulties with smaller test trajectory lengths such as 50–100 due to training data imbalance. Overall, the findings highlight the promise of hybrid deep learning and similarity-based approaches for RUL prediction. However, potential advancements such as hyperparameter optimisation and data augmentation still hold potential for further improvements.

Basin-informed flood frequency analysis using deep learning exhibits consistent projected regional patterns over CONUS

Climate change poses a significant threat to flood-prone areas by altering precipitation patterns and the water cycle. Here, we analyzed the impact of climate change on future flood trends. We trained a Long Short-Term Memory (LSTM) model to estimate long term discharge at 638 river sites over contiguous United States (CONUS) based on inputs from the gridMET meteorological datasets, and downscaled and bias-corrected Coupled Model Intercomparison Project 5 (CMIP5) projections. Our results indicate that the LSTM model can replicate observed discharge with reliable accuracy. The projected changes in flood magnitude for the 10-year and 100-year return periods reveal consistent geographical patterns robust across climate models, with increasing trends of approximately + 10 to + 40% in the East and West coastal regions and decreasing trends of about − 10 to − 30% in the Southwestern areas. The regions exhibiting an increasing flood trend are likely driven by an increase in total seasonal extreme precipitation and changes in the timing and amount of peak flow. In contrast, the decreasing flood trends result from a significant reduction in snowpack. To support adaptation planning, we developed an interactive map providing the historical and projected flood changes for 10- and 100-year floods across the 638 selected basins over CONUS.

Using different machine learning models for address resolution protocol spoofing attack detection in software-defined network architecture

Purpose The purpose of the study concludes detecting address resolution protocol (ARP) Spoofing attack in software-defined network (SDN) architecture meanwhile using different machine learning models to evaluate their effectiveness. Design/methodology/approach This research originates from building a SDN topology and researching into its changes under ARP Spoofing attack. Based on that, the authors propose four features which show obvious abnormalities in network under attack stage. The data collected from SDN controller is used to build a data set, which is then put into different machine learning models, which are: Artificial Neuron network (ANN), Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), CNN-LSTM and Gated Recurrent Unit (GRU). Findings After applying this proposal in simulation and experimental environments, they achieve impressive performance metrics. In simulation environment, the GRU model stands out with the highest accuracy of 98.94%. In real environments, the CNN-LSTM model leads with a recall of 98.38% and an F1-Score of 98.57%, while the LSTM model has the highest precision (98.8%). The GRU model also performs strongly in real scenarios with a high accuracy of 97.65%. ANN, despite its reliability, struggles with lower recall and F1-Score across both environments. Originality/value This analysis emphasizes the importance of the proposed features when applied to different models and their high potential to conduct in practical environment.

Mobile Application for Continuous Recognition and Classification of Sign Language Images through Deep Learning

Throughout the world, sign languages (SL) present significant challenges for effective communication in everyday environments and technological applications. In the field of SL recognition (SLR) using artificial intelligence (AI), two approaches have been developed: isolated SLR (ISLR) and continuous SLR (CSLR). To overcome the limitations of CSLR in SL, we developed a mobile application that integrates an AI-based algorithm in Python, designed to capture and analyze sign sequences through the device’s camera. The application facilitates the creation of a continuous database containing 14 dynamic signs, with 240 videos per sign, resulting in a total of 3360 videos and 50,400 frames. We used a neural network model based on the long short-term memory (LSTM) architecture to improve accuracy in sign identification and promote inclusive communication in digital environments. The model achieved 99.80% accuracy during training and 99.40% in testing, with overall accuracy, recall, and F1-score metrics above 99%. These results evidence the effectiveness of the mobile application and the LSTM model in recognizing, classifying, and translating basic SLP utterances in real time, demonstrating its ability to generalize and avoid overfitting and contributing to more inclusive and accessible communication.

Data driven modeling of TiO2 PVP nanofiber diameter using LSTM and regression for enhanced functional performance

The prospective utilization of electrospun nanofibers across diverse fields has elicited substantial scientific attention. Nevertheless, managing their diameter remains problematic due to the intricate interactions among electrospinning variables. This research explores the application of Long Short-Term Memory (LSTM) networks and multiple regression models to forecast the diameters of Titanium Dioxide (TiO₂) and Polyvinyl pyrrolidone (PVP) nanofibers, facilitating improved process regulation and enhancement. TiO₂ + PVP nanofibers were fabricated under diverse conditions, including changes in applied voltage, solution concentration, and distance between tip to collector. The acquired data underwent analysis using LSTM and regression models to assess their predictive capabilities. The outcomes revealed that both approaches effectively estimated nanofiber diameters; however, the regression model surpassed LSTM with a lower error rate of 0.077% compared to 0.305%. This indicates that while LSTM captures nonlinear relationships, conventional regression models yield more precise predictions in this scenario. These findings underscore the potential of machine learning in advancing electrospinning technology by minimizing trial-and-error experiments and boosting nanofiber production efficiency. The incorporation of artificial intelligence-driven modeling into the electrospinning process sets the stage for more accurate control over fiber morphology, resulting in enhanced material properties and expanded applications in biomedical, environmental, and energy sectors.

Comparing the Effectiveness of Deep Learning Approaches for Charging Time Prediction in Electric Vehicles: Kocaeli Example

The aim of this study is to compare the performance of various deep learning models in predicting the arrival and charging time of an electric vehicle at a charging station. The objective is to identify the most precise model capable of predicting the time from the driver’s location to the arrival at the charging station. Initially, an effort was made to ascertain which model type offers superior prediction accuracy, characterized by low error rates and high success scores. To this end, the study examined the prediction capabilities of the LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), Prophet, and ARIMA (Autoregressive Integrated Moving Average) models, which were trained with historical data using various metrics. These metrics included the R2 (R-squared)success metric, MAE (Mean Absolute Error) and MSE (Mean Squared Error) error metrics, DTW (Dynamic Time Warping) metric scores, and model selection. In MAE metric scores, GRU has an error rate of 0.38, LSTM 0.50, Prophet 0.61, and ARIMA 2.31. For MSE metric scores, GRU has an error rate of 2.92, LSTM 3.05, Prophet 4.21, and ARIMA 65.54. In R2 metric scores, GRU has a success rate of 0.99, LSTM 0.99, Prophet 0.13, and ARIMA −2.19. In DTW metric scores, GRU has a distance ratio of 125.5, LSTM 126.9, Prophet 185.9, and ARIMA 454.9. Based on these score values, it was decided that the GRU model made the most accurate time estimation. After observing the superior performance of the GRU model, the time prediction capability of this model is demonstrated through an interface program for charging stations in the province of Kocaeli, which serves as the model’s real-world application area. The main contribution of this article is that LSTM, GRU, Prophet, and ARIMA deep learning approaches, which are preferred in many studies, are used for the first time in the process of estimating electric vehicle charging time.

A Time Series Data Classification Method for Gesture Recognition Based on LSTM and Attention Mechanism

This paper addresses the issues of weak model transferability and poor environmental adaptability in cross-domain gesture recognition within wireless sensing technology. A time-series data classification model integrating Long Short-Term Memory (LSTM) and attention mechanism (W-LSTM+A) is proposed. By introducing a feature selection weight matrix to reconstruct the LSTM gating mechanism and combining a dynamic attention allocation strategy, the models ability to capture key spatiotemporal features in channel state information is significantly enhanced. Experiments based on a WiFi signal dataset collected in a real office environment compared the performance of CNN, LSTM, and LSTM+A models. The results show that the LSTM+A model achieved a test accuracy of 87.3% after 200 training epochs, significantly outperforming CNNs 81.9%. Although the LSTM model had a higher final accuracy, its convergence speed was significantly slower than that of the LSTM+A model. Further analysis indicates that the attention mechanism, by strengthening key time-step features, enables the model to quickly capture effective patterns in the early stages of training. However, due to limited sample size, its potential has not been fully realized. This study provides new solutions for the cross-scene adaptability of wireless sensing technology and has application value in smart homes, health monitoring, and other fields.

Assessing dengue forecasting methods: a comparative study of statistical models and machine learning techniques in Rio de Janeiro, Brazil

BackgroundDengue is a mosquito-borne viral disease that poses a significant public health threat in tropical and subtropical regions worldwide. Accurate forecasting of dengue outbreaks is crucial for effective public health planning and intervention. This study aims to assess the predictive performance and computational efficiency of a number of statistical models and machine learning techniques for dengue forecasting, both with and without the inclusion of climate factors, to inform the design of dengue surveillance systems.MethodsThe dengue forecasting methods comparison in this study considers dengue cases in Rio de Janeiro, Brazil, as well as climate factors known to affect disease transmission. Employing a dynamic window approach, various statistical methods and machine learning techniques were used to generate weekly forecasts at several time horizons. Error measures, uncertainty intervals, and computational efficiency obtained with each method were compared. Statistical models considered were Autoregressive (AR), Moving Average (MA), Autoregressive Integrated Moving Average (ARIMA), and Exponential Smoothing State Space Model (ETS). In addition, models incorporating temperature and humidity as covariates, such as Vector Autoregression (VAR) and Seasonal ARIMAX (SARIMAX), were employed. Machine learning techniques evaluated were Random Forest, XGBoost, Support Vector Machine (SVM), Long–Short-Term Memory (LSTM) networks, and Prophet. Ensemble approaches that integrated the top performing methods were also considered. The evaluated methods also incorporated lagged climatic variables to account for delayed effects.ResultsAmong the statistical models, ARIMA demonstrated the best performance using only historical case data, while SARIMAX significantly improved predictive accuracy by incorporating climate covariates. In general, the LSTM model, particularly when combined with climate covariates, proved to be the most accurate machine learning model, despite being slower to train and predict. For long-term forecasts, Prophet with climate covariates was the most effective. Ensemble models, such as the combination of LSTM and ARIMA, showed substantial improvements over individual models.ConclusionsThis study demonstrates the strengths and limitations of various methods for dengue forecasting across multiple timeframes. It highlights the best-performing statistical and machine learning methods, including their computational efficiency, underscoring the significance of machine learning techniques and the integration of climate covariates to improve forecasts. These findings offer valuable insights for public health officials, facilitating the development of dengue surveillance systems for more accurate forecasting and timely allocation of resources to mitigate dengue outbreaks.

Mapping near-real-time soil moisture dynamics over Tasmania with transfer learning

Abstract. Soil moisture, an essential parameter for hydroclimatic studies, exhibits considerable spatial and temporal variability, which complicates its mapping at high spatiotemporal resolutions. Although current remote sensing products offer global estimates of soil moisture at fine temporal resolutions, they do so at a coarse spatial resolution. Deep learning (DL) techniques have recently been employed to produce high-resolution maps of various soil properties; however, these methods require substantial training data. This study sought to map daily soil moisture across Tasmania, Australia, at an 80 m resolution using a limited set of training data. We assessed three modeling strategies: DL models calibrated using an Australian dataset (51 411 observation points), models calibrated using the Tasmanian dataset (9825 observation points), and a transfer learning technique that transferred information from the Australian models to Tasmania using region-specific data. We also evaluated two DL approaches, i.e., multilayer perceptron (MLP) and long short-term memory (LSTM). The models included the Soil Moisture Active Passive (SMAP) dataset, weather data, an elevation map, land cover, and multilevel soil property maps as inputs to generate soil moisture at the surface (0–30 cm) and subsurface (30–60 cm) layers. Results showed that (1) models calibrated from the Australian dataset performed worse than Tasmanian models regardless of the type of DL approaches; (2) Tasmanian models, calibrated solely using local data, resulted in shortcomings in predicting soil moisture; and (3) transfer learning exhibited remarkable performance improvements (error reductions of up to 45 % and a 50 % increase in correlation) and resolved the drawbacks of the two previous models. The LSTM models with transfer learning had the highest overall performance with an average mean absolute error (MAE) of 0.07 m3 m−3 and a correlation coefficient (r) of 0.77 across stations for the surface layer as well as MAE=0.07m3m-3 and r=0.69 for the subsurface layer. The fine-resolution soil moisture maps captured the detailed landscape variation as well as temporal variation according to four distinct seasons in Tasmania. The models were then applied to generate daily soil moisture maps of Tasmania, integrated into a near-real-time monitoring system to assist agricultural decision-making.

SapFlower: an automated tool for sap flow data preprocessing, gap-filling, and analysis using deep learning.

Sap flow, a critical process in plant water use and ecosystem water cycles, is often measured using thermal dissipation probes (TDP) due to their ease of installation and continuous data collection. However, sap flow data frequently include noise, outliers, and gaps, creating challenges for analysis and requiring substantial manual processing. We developed SapFlower, a tool that automates data preprocessing, model training, gap-filling, sapwood area scaling and modeling, and water use analysis. It integrates autocleaning, machine learning and deep learning models (e.g. random forest, Gaussian process regression, long short-term memory (LSTM), bidirectional LSTM (BiLSTM)), and efficient workflows to process sap flow data. SapFlower can remove over 90% of noisy data while preserving legitimate variations and achieve high accuracy in gap-filling based on user-determined parameters. Random forest, LSTM, and BiLSTM models reduced root mean square error to 10% or less for long-term gaps. Model training and prediction can be performed efficiently within seconds. SapFlower significantly enhances the efficiency and accessibility of TDP data analysis by automating complex tasks, enabling researchers without programming expertise to employ advanced techniques. Future improvements will focus on species-specific corrections for TDP and support for additional measurement methods. SapFlower is openly available on GitHub (https://github.com/JiaxinWang123/SapFlower) and Zenodo (doi: 10.5281/zenodo.13665919).

Predicting Cryptocurrency Prices with a Hybrid ARIMA and LSTM Model

Cryptocurrencies have attracted significant attention from investors, regulators and the media since their emergence. In a world where digital advancements are increasingly included in everyday relations, studying the behaviour of cryptocurrencies and their impact on financial markets becomes a necessity. This paper introduces a comparative analysis towards a hybrid model combining classical and modern methods for predicting cryptocurrency prices. This study deals with everyday recordings of 10 cryptocurrencies that represent different technological innovations and use cases. Studying these cryptocurrencies can help understand volatility, volumes and price movements. We aim to develop a time series statistical model and to study the effectiveness of deep learning (DL) models, specifically long short-term memory (LSTM) model and the autoregressive integrated moving average (ARIMA) model, for predicting cryptocurrency prices accurately and forecasting stationary data. Combining ARIMA and LSTM, we managed to obtain a high value of R² for Binance Coin (BNB) cryptocurrency (0.936) with an average R² for all evaluated cryptocurrencies of 0.6555.

Anomaly Detection in IoT-Connected Medical Devices Using Machine Learning for Disease Monitoring

More and more, the Internet of Things (IoT) is being used in healthcare. This has made disease tracking much better by letting medical gadgets and apps send data in real time. But it's still hard to make sure that this data is correct and complete, especially when there are oddities that can happen because of broken devices, online threats, or strange physical situations. This paper discusses a machine learning approach for locating unusual items in IoT-connected medical equipment. The aim is to increase the dependability of disease surveillance systems. The proposed approach finds unusual patterns in streams of body data distinct from one another using unsupervised and semi-supervised learning models such as Isolation Forest, Autoencoders, and Long Short-Term Memory (LSTM) networks. The system architecture is suitable for real-time healthcare applications as it can be implemented on edge, fog, and cloud platforms. With an F1-score of 0.86 and an AUC of 0.91, the LSTM model was the most accurate based on testing utilising both fake and actual datasets. It outperformed conventional techniques such as k-means clustering and Z-score. Two graphical techniques that indicate how well the intended system functions are ECG anomaly detection plots and ROC curves. A flexible and explainable machine learning process, context-aware anomaly scores with EHR integration, and new ideas about how to make models more general and how to balance computing needs are some of the most important advances. These results show that intelligent anomaly detection systems can help with early action, cut down on fake alarms, and make smart healthcare settings safer for patients.

Bitcoin Price Prediction Using LSTM-Based Deep Learning Techniques in Python: A Data-Driven Approach

This study examines the cryptocurrency markets that have experienced exponential growth over the past decade, with Bitcoin emerging as the most dominant and widely traded digital currency. However, the inherent volatility and non-linear behavior of Bitcoin prices pose significant challenges in forecasting. This study aims to investigate the efficacy of deep learning models, particularly Long Short-Term Memory (LSTM) networks, in predicting Bitcoin prices using historical data. The research utilizes daily price data, cleansed and normalized, and applies LSTM-based neural networks to capture the temporal dynamics of Bitcoin’s market behavior. The model was trained using Python-based frameworks such as Keras and TensorFlow, and its performance was evaluated using standard error metrics like Mean Squared Error (MSE) and Root Mean Squared Error (RMSE). The results indicate that the LSTM model is capable of accurately predicting future trends in Bitcoin prices, outperforming traditional linear forecasting methods. This research highlights the growing relevance of deep learning in the financial sector, especially in volatile and data-rich environments like cryptocurrency. Keywords: Bitcoin; Deep Learning; LSTM; Price Prediction; Time-Series Forecasting; Python; Cryptocurrency

Multivariate Time Series Models for Soil Nutrient and Yield Prediction in Site Specific Integrated Nutrient Management

ABSTRACT In agriculture, precise prediction of soil nutrient dynamics and crop yield under Site-Specific Integrated Nutrient Management (SSNM) is essential for optimizing resource efficiency and enhancing sustainability. SSNM can be integrated with LSTM, ARIMA, or SARIMA models to optimize nutrients based on real-time or forecasted data. While SSNM focuses on personalized nutrient requirements for specific sites, these time series models, like LSTM (Long Short-Term Memory), ARIMA (AutoRegressive Integrated Moving Average), and SARIMA (Seasonal ARIMA), can be used to predict environmental factors such as weather patterns, crop growth conditions, and soil nutrient levels over time. This study applies advanced multivariate time series models, including Long Short-Term Memory (LSTM) networks, AutoRegressive Integrated Moving Average (ARIMA), and Seasonal ARIMA (SARIMA) to establish their complementary role in SSNM implementation. These models help to predict the dynamics of soil nutrients such as nitrogen (N), phosphorus (P), and potassium (K) along with crop yield over a 25-year (2025–2050). LSTM demonstrated superior predictive accuracy for soil nutrient dynamics (R2 = 0.99), while ARIMA outperformed for crop yield predictions (R2 = 0.58), effectively capturing temporal fluctuations. However, the modest performance of ARIMA for yield prediction due to exogenous variables, seasonal and structural trends. The results highlight the strengths of these models in supporting SSNM by enabling accurate forecasts of key parameters for data-driven decision-making. This research underscores the importance of integrating dynamic modeling approaches with SSNM to enhance agricultural productivity, improve resource use efficiency, and support sustainable crop management strategies.

Viral particle prediction in wastewater treatment plants using nonlinear lifelong learning models

Predicting new unseen data using only wastewater process inputs remains an open challenge. This paper proposes lifelong learning approaches that integrate long short-term memory (LSTM), gated recurrent unit (GRU) and tree-based machine learning models with knowledge-based dictionaries for real-time viral prediction across various wastewater treatment plants (WWTPs) in Saudi Arabia. Limited data prompted the use of a Wasserstein generative adversarial network to generate synthetic data from physicochemical parameters (e.g., pH, chemical oxygen demand, total dissolved solids, total suspended solids, turbidity, conductivity, NO2-N, NO3-N, NH4-N), virometry, and PCR-based methods. The input features and predictors are combined into a coupled dictionary learning framework, enabling knowledge transfer for new WWTP batches. We tested the framework for predicting total virus, adenovirus, and pepper mild mottle virus from WWTP stages, including conventional activated sludge, sand filter, and ultrafiltration effluents. The LSTM and GRU models adapted well to new data, maintaining robust performance. Tests on total viral prediction across four municipal WWTPs in Saudi Arabia showed the lifelong learning model’s value for adaptive viral particle prediction and performance enhancement.

Rainfall Prediction in Jayapura City Area Using Long Short-Term Memory

Jayapura, one of Indonesia’s major fishing cities, relies heavily on accurate weather predictions to ensure the safety of its fishermen, particularly due to its significant tuna and skipjack production. This study aims to improve rainfall forecasting in Jayapura using a Long Short-Term Memory (LSTM) model, a type of artificial neural network designed for time series prediction. Accurate rainfall forecasts are crucial for reducing the risks fishermen face at sea due to sudden weather changes. Daily data from the Meteorological Station in Dok II Jayapura was collected and processed to train the LSTM model, incorporating variables such as TAVG (average temperature), RH_AVG (average relative humidity), FF_AVG (average wind speed), Pressure (air pressure), and Wind_Gust (wind gust). The model’s performance was evaluated using Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE), yielding low values of 0.0542 and 0.0847, respectively, indicating high prediction accuracy. The MAE reflects the average magnitude of errors, while the RMSE highlights the model’s sensitivity to larger deviations, both supporting the reliability of the LSTM approach. The findings demonstrate that LSTM models can effectively forecast rainfall in Jayapura, providing valuable information that helps fishermen plan their activities more safely and efficiently. The study concludes that LSTM is a robust tool for rainfall prediction, and the inclusion of additional meteorological variables has proven to enhance accuracy. Further research is recommended to explore other factors to improve prediction reliability.

Development of an In-Vehicle Intrusion Detection Model Integrating Federated Learning and LSTM Networks

Introduction: Ensuring vehicular cybersecurity is a critical challenge due to the increasing connectivity of modern vehicles, and traditional centralised learning approaches for intrusion detection pose significant privacy risks, as they require sensitive data to be shared from multiple vehicles to a central server. Objective: The aim of this study is therefore to develop an in-vehicle intrusion detection system (IVIDS) that integrates federated learning (FL) with neural networks, enabling decentralised and privacy-preserving detection of cyberattacks in vehicular networks. The proposed system extends previous research by detecting a broader range of attacks (eight types) and exploring different deep learning architectures. Methods: This study employs an extended version of the publicly available VeReMi dataset to train and evaluate multiple neural network architectures, including Multilayer Perceptrons (MLPs), Gated Recurrent Units (GRUs), and Long Short-Term Memory (LSTM) networks. Federated learning is utilised to enable collaborative model training across multiple vehicles without sharing raw data. Various data preprocessing techniques and differential privacy mechanisms are also explored. Results and Conclusions: The experimental results demonstrate that LSTM networks outperform both MLP and GRU architectures in classifying vehicular cyberattacks. The best LSTM model, trained with two previous message lags and standard normalisation, achieved a classification accuracy of 96.75% in detecting eight types of attacks, surpassing previous studies, and demonstrating the potential of applying neural networks designed to work with time series data.

A Comprehensive Study of Deep Learning Models for Intrusion Detection in IoT Devices

The Internet of Things (IoT) has revolutionized how people interact with the world, but the increasing complexity of cyberattacks poses significant challenges in detecting intrusions. Failure to prevent intrusions can compromise IoT security services, including data confidentiality, integrity, and availability. For this reason, this study employs four deep learning models: A Deep Neural Networks (DNN), a Convolutional Neural Network (CNN), a Recurrent Neural Network (RNN), and a Long-Short-Term Memory (LSTM) network. The multiclassification performance of each model was evaluated using the Bot-IoT dataset. This study also addresses the bias towards the DDoS/DoS category in the Bot-IoT dataset, using the SMOTE technique to mitigate overfitting. The LSTM model achieved an excellent balance between performance and efficiency, outperforming state-of-the-art deep learning Intrusion Detection System (IDS) approaches on the same dataset, achieving a multiclass classification accuracy of 99.97%.