PeerJ Computer Science最新文献

Big data analytics for clinical decision-making has been proposed for various clinical sectors because clinical decisions are more evidence-based and promising. Healthcare data is so vast and readily available that big data analytics has completely transformed this sector and opened up many new prospects. The smart sensor-based big data analysis recommendation system has significant privacy and security concerns when using sensor medical images for suggestions and monitoring. The danger of security breaches and unauthorized access, which might lead to identity theft and privacy violations, increases when sending and storing sensitive medical data on the cloud. Our effort will improve patient care and well-being by creating an anomaly detection system based on machine learning specifically for medical images and providing timely treatments and notifications. Current anomaly detection methods in healthcare systems, such as artificial intelligence and big data analytics-intracerebral hemorrhage (AIBDA-ICH) and parallel conformer neural network (PCNN), face several challenges, including high resource consumption, inefficient feature selection, and an inability to handle temporal data effectively for real-time monitoring. Techniques like support vector machines (SVM) and the hidden Markov model (HMM) struggle with computational overhead and scalability in large datasets, limiting their performance in critical healthcare applications. Additionally, existing methods often fail to provide accurate anomaly detection with low latency, making them unsuitable for time-sensitive environments. We infer the extraction, feature selection, attack detection, and data collection and processing procedures to anticipate anomaly inpatient data. We transfer the data, take care of missing values, and sanitize it using the pre-processing mechanism. We employed the recursive feature elimination (RFE) and dynamic principal component analysis (DPCA) algorithms for feature selection and extraction. In addition, we applied the Auto-encoded genetic recurrent neural network (AGRNN) approach to identify abnormalities. Data arrival rate, resource consumption, propagation delay, transaction epoch, true positive rate, false alarm rate, and root mean square error (RMSE) are some metrics used to evaluate the proposed task.

Traditional disease retrieval and localization for plant leaves typically demand substantial human resources and time. In this study, an intelligent approach utilizing deep hash convolutional neural networks (DHCNN) is presented to address these challenges and enhance retrieval performance. By integrating a collision-resistant hashing technique, this method demonstrates an improved ability to distinguish highly similar disease features, achieving over 98.4% in both precision and true positive rate (TPR) for single-plant disease retrieval on crops like apple, corn and tomato. For multi-plant disease retrieval, the approach further achieves impressive Precision of 99.5%, TPR of 99.6% and F-score of 99.58% on the augmented PlantVillage dataset, confirming its robustness in handling diverse plant diseases. This method ensures precise disease retrieval in demanding conditions, whether for single or multiple plant scenarios.

Security requirements are considered one of the most important non-functional requirements of software. The CIA (confidentiality, integrity, and availability) triad forms the basis for the development of security systems. Each dimension is expressed as having many security requirements that should be designed, implemented, and tested. However, requirements are written in a natural language and may suffer from ambiguity and inconsistency, which makes it harder to distinguish between different security dimensions. Recognizing the security dimensions in a requirements document should facilitate tracing the requirements and ensuring that a dimension has been implemented in a software system. This process should be automated to reduce time and effort for software engineers. In this paper, we propose to classify the security requirements into CIA triads using Term frequency-inverse document frequency and sentence-transformer embedding as two different technologies for feature extraction. For both techniques, we developed five models by using five well-known machine learning algorithms: (1) support vector machine (SVM), (2) K-nearest neighbors (KNN), (3) Random Forest (RF), (4) gradient boosting (GB), and (5) Bernoulli Naive Bayes (BNB). Also, we developed a web interface that facilitates real-time analysis and classifies security requirements into CIA triads. Our results revealed that SVM with the sentence-transformer technique outperformed all classifiers by 87% accuracy in predicting a type of security dimension.

To address the issues of data sparsity, scalability, and cold start in the traditional teaching resource recommendation process, this paper presents an enhanced collaborative filtering (CF) recommendation algorithm incorporating a time decay (TD) function. By aligning with the human memory forgetting curve, the TD function is employed as a weighting factor, enabling the calculation of similarity and user preferences constrained by the TD, thus amplifying the weight of user interest over a short period and achieving the integration of short-term and long-term interests. The results indicate that the RMSE of the proposed combined recommendation algorithm (TD-CF) is only 8.95 when the number of recommendations reaches 100, which is significantly lower than the comparison model, which exhibits higher accuracy across different recommended items, effectively utilizing music teaching resources and user characteristics to deliver more precise recommendations.

Climate change has become a major source of concern to the global community. The steady pollution of the environment including our waters is gradually increasing the effects of climate change. The disposal of plastics in the seas alters aquatic life. Marine plastic pollution poses a grave danger to the marine environment and the long-term health of the ocean. Though technology is also seen as one of the contributors to climate change many aspects of it are being applied to combat climate-related disasters and to raise awareness about the need to protect the planet. This study investigated the amount of pollution in marine and undersea leveraging the power of artificial intelligence to identify and categorise marine and undersea plastic wastes. The classification was done using two types of machine learning algorithms: two-step clustering and a fully convolutional network (FCN). The models were trained using Kaggle's plastic location data, which was acquired in situ. An experimental test was conducted to validate the accuracy and performance of the trained models and the results were promising when compared to other conventional approaches and models. The model was used to create and test an automated floating plastic detection system in the required timeframe. In both cases, the trained model was able to correctly identify the floating plastic and achieved an accuracy of 98.38%. The technique presented in this study can be a crucial instrument for automatic detection of plastic garbage in the ocean thereby enhancing the war against marine pollution.

We present the package prosper_nn, that provides four neural network architectures dedicated to time series forecasting, implemented in PyTorch. In addition, prosper_nn contains the first sensitivity analysis suitable for recurrent neural networks (RNN) and a heatmap to visualize forecasting uncertainty, which was previously only available in Java. These models and methods have successfully been in use in industry for two decades and were used and referenced in several scientific publications. However, only now we make them publicly available on GitHub, allowing researchers and practitioners to benchmark and further develop them. The package is designed to make the models easily accessible, thereby enabling research and application in various fields like demand and macroeconomic forecasting.

Motion blur is a problem that degrades the visual quality of images for human perception and also challenges computer vision tasks. While existing studies mostly focus on deblurring algorithms to remove uniform blur due to their computational efficiency, such approaches fail when faced with non-uniform blur. In this study, we propose a novel algorithm for motion deblurring that utilizes an adaptive mesh-grid approach to manage non-uniform motion blur with a focus on reducing the computational cost. The proposed method divides the image into a mesh-grid and estimates the blur point spread function (PSF) using an inertial sensor. For each video frame, the size of the grid cells is determined adaptively according to the in-frame spatial variance of blur magnitude which is a proposed metric for the blur non-uniformity in the video frame. The adaptive mesh-size takes smaller values for higher variances, increasing the spatial accuracy of the PSF estimation. Two versions of the adaptive mesh-size algorithm are studied, optimized for either best quality or balanced performance and computation cost. Also, a trade-off parameter is defined for changing the mesh-size according to application requirements. The experiments, using real-life motion data combined with simulated motion blur demonstrate that the proposed adaptive mesh-size algorithm can achieve 5% increase in PSNR quality gain together with a 19% decrease in computation time on the average when compared to the constant mesh-size method.

At CRYPTO 2019, Gohr pioneered the application of deep learning to differential cryptanalysis and successfully attacked the 11-round NSA block cipher Speck32/64 with a 7-round and an 8-round single-key differential neural distinguisher. Subsequently, Lu et al. (DOI 10.1093/comjnl/bxac195) presented the improved related-key differential neural distinguishers against the SIMON and SIMECK. Following this work, we provide a framework to construct the enhanced related-key differential neural distinguisher for SIMON and SIMECK. In order to select input differences efficiently, we introduce a method that leverages weighted bias scores to approximate the suitability of various input differences. Building on the principles of the basic related-key differential neural distinguisher, we further propose an improved scheme to construct the enhanced related-key differential neural distinguisher by utilizing two input differences, and obtain superior accuracy than Lu et al. for both SIMON and SIMECK. Specifically, our meticulous selection of input differences yields significant accuracy improvements of 3% and 1.9% for the 12-round and 13-round basic related-key differential neural distinguishers of SIMON32/64. Moreover, our enhanced related-key differential neural distinguishers surpass the basic related-key differential neural distinguishers. For 13-round SIMON32/64, 13-round SIMON48/96, and 14-round SIMON64/128, the accuracy of their related-key differential neural distinguishers increases from 0.545, 0.650, and 0.580 to 0.567, 0.696, and 0.618, respectively. For 15-round SIMECK32/64, 19-round SIMECK48/96, and 22-round SIMECK64/128, the accuracy of their neural distinguishers is improved from 0.547, 0.516, and 0.519 to 0.568, 0.523, and 0.526, respectively.

Industrial organizations are turning to recommender systems (RSs) to provide more personalized experiences to customers. This technology provides an efficient solution to the over-choice problem by quickly combing through large amounts of information and supplying recommendations that fit each user's individual preferences. It is quickly becoming an integral part of operations, as it yields successful and convenient results. This research provides an enhanced integrated fuzzy logic-based deep learning technique (EIFL-DL) for recent industrial challenges. Extracting useful insights and making appropriate suggestions in industrial settings is difficult due to the fast development of data. Traditional RSs often struggle to handle the complexity and uncertainty inherent in industrial data. To address these limitations, we propose an EIFL-DL framework that combines fuzzy logic and deep learning techniques to enhance recommendation accuracy and interpretability. The EIFL-DL framework leverages fuzzy logic to handle uncertainty and vagueness in industrial data. Fuzzy logic enables the modelling of imprecise and uncertain information, and the system is able to capture nuanced relationships and make more accurate recommendations. Deep learning techniques, on the other hand, excel at extracting complex patterns and features from large-scale data. By integrating fuzzy logic with deep learning, the EIFL-DL framework harnesses the strengths of both approaches to overcome the limitations of traditional RSs. The proposed framework consists of three main stages: data preprocessing, feature extraction, and recommendation generation. In the data preprocessing stage, industrial data is cleaned, normalized, and transformed into fuzzy sets to handle uncertainty. The feature extraction stage employs deep learning techniques, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs), to extract meaningful features from the preprocessed data. Finally, the recommendation generation stage utilizes fuzzy logic-based rules and a hybrid recommendation algorithm to generate accurate and interpretable recommendations for industrial applications.

Purpose: This study aims to address the limitations of traditional data processing methods in predicting agricultural product prices, which is essential for advancing rural informatization to enhance agricultural efficiency and support rural economic growth.

Methodology: The RL-CNN-GRU framework combines reinforcement learning (RL), convolutional neural network (CNN), and gated recurrent unit (GRU) to improve agricultural price predictions using multidimensional time series data, including historical prices, weather, soil conditions, and other influencing factors. Initially, the model employs a 1D-CNN for feature extraction, followed by GRUs to capture temporal patterns in the data. Reinforcement learning further optimizes the model, enhancing the analysis and accuracy of multidimensional data inputs for more reliable price predictions.

Results: Testing on public and proprietary datasets shows that the RL-CNN-GRU framework significantly outperforms traditional models in predicting prices, with lower mean squared error (MSE) and mean absolute error (MAE) metrics.

Conclusion: The RL-CNN-GRU framework contributes to rural informatization by offering a more accurate prediction tool, thereby supporting improved decision-making in agricultural processes and fostering rural economic development.