Evaluation Metrics Research Articles

Predictions of Peak Discharge of Dam Failures Based on the Combined GA and BP Neural Networks

In this paper, the Artificial Neural Network (ANN) was utilized to predict the peak discharge of dam failures, which was based on the combined Genetic Algorithm (GA) and Back Propagation (BP) neural network. The dataset comprises 40 samples from self-conducted experiments and available literature. To compare the efficiency of the suggested approach, three evaluation metrics, including the coefficient of determination (R2), the root mean square error (RMSE) and the mean absolute error (MAE), were analyzed for both the BP neural network and the GA-BP neural network. The findings suggest that (1) The prediction accuracy of the GA-BP was better than that of the BP; and (2) Compared to BP, GA-BP demonstrated a 9.07% average improvement in R2, a 57.36% average reduction in MAE, and a 57.53% average reduction in RMSE. In addition, the results of GA-BP and semi-empirical formulas were compared and the effect of three parameters on the peak discharge was analyzed. The results showed that the GA-BP model could effectively predict the peak discharge of dam failures.

A transformer-based approach to Nigerian Pidgin text generation

AbstractThis paper describes the development of a transformer-based text generation model for Nigerian Pidgin also known as Naijá, a popular language in West Africa. Despite its wide use, Nigerian Pidgin remains under-resourced, particularly in areas related to text generation and natural language processing. These difficulties are primarily due to technological constraints rather than the language’s fundamental attributes. There is currently a demand for Nigerian Pidgin-specific solutions because it is used in everyday communication and has a unique linguistic blend. This paper aims to close this gap by exploring the application of state-of-the-art transformer technology to develop a text generation model for Nigerian Pidgin. This work uses the public Afriberta-corpus dataset to optimize the Generative Pre-trained Transformer (GPT-2) model across a sizeable dataset. The performance evaluators, BLEU and Perplexity metrics provide a detailed breakdown of the model’s text quality and predictive accuracy. Despite the difficulties caused by a limited amount of training data, preliminary evaluations show that the model can generate coherent Nigerian Pidgin text. The performance evaluation yielded perplexity scores of 43.56 for variable target reference length and 43.26 for fixed text length. BLEU scores of 0.15 for fixed max length and 0.56 for variable reference target length. This highlights the quality of generated text and the significant improvement when the generated text length is aligned with the reference target. Our work was benchmarked against African American Vernacular (AAVE) revealing that BLEU scores for AAVE are significantly lower than those for Standard American English, with BLEU given as 0.26. Our Nigerian Pidgin model, with a BLEU score of 0.56, shows a better performance. However, both results suggest that both dialects are challenging for language models. Leveraging the pre-trained transformer-based language model and evaluation metrics, we showcase the model’s capacity for coherent Nigerian Pidgin text generation. For future research, the research work can serve as a good foundation for advancement and progress in the Nigerian Pidgin language generation and other low-resource languages.

Point cloud upsampling network based on pyramid pooling and self-attention mechanism

For the problems of irregular point cloud data and non-unique input size, a point-cloud upsampling network (PPSA-PU) based on Pyramid Pooling and Self-Attention mechanism is proposed. Multi-scale features are first extracted from the input point cloud using dense map convolution with different expansion rates, and the pyramid pooling module and self-attention mechanism are added. By applying the same pooling operation to regions of different sizes and numbers, the network is made to accept inputs of arbitrary size. Self-attention is used to weight the features to establish global correlation and improve the feature extraction capability. The features are then extended by the Self-Attention NodeShuffle (SA-NS) module, which uses a graph convolution layer to fuse neighbour information and learn new point features from the latent space. Finally, the coordinate reconstruction module is used to generate a dense up-sampled point cloud. Using the existing dataset to train and test the network and compare it with the existing up-sampling network, the experimental results show that PPSA-PU is better in terms of evaluation metrics and performance, and has better generalisation and robustness.

How accurately can supervised machine learning model predict a targeted psychiatric disorder?

BackgroundHoarding disorder (HD) is characterized by a compulsion to collect belongings, and to experience significant distress when parting from them. HD is often misdiagnosed for several reasons. These include patient and family lack of recognition that it is a psychiatric disorder and professionals’ lack of relevant expertise with it. This study evaluates the ability of a supervised machine learning (ML) model to match the diagnostic skills of psychiatrists when presented with equivalent information pertinent to symptoms of HD.MethodsFive hundred online participants were randomly recruited and completed the Hoarding Rating Scale-Self Report (HRS-SR) and the Generalized Anxiety Disorder 7-item (GAD-7) scale. Responses to the questionnaires were read by an ML model. Responses to the HRS-SR were then converted into anonymized, random-equivalent texts. Each of these individual texts was presented in random order to two experienced psychiatrists who were independently asked for a provisional diagnosis - e.g.; the presence or absence of HD. In case of disagreement between the two assessors, a third psychiatrist broke the tie. A decision tree classification model was employed to predict clinical HD using self-report data from two psychological tests, the HRS-SR and GAD-7. The target variable was whether a participant had clinical HD, while the predictive variables were the continuous scores from the HRS-SR and GAD-7 tests. The model’s performance was evaluated using a confusion matrix, which compared the observed diagnoses with the predicted diagnoses to assess accuracy.ResultsAccording to the psychiatrists, approximately 10% of the participants fulfilled DSM-5 diagnostic criteria for HD. 93% of the clinician-identified cases were identified by the ML model based on HRS-SR and GAD-7 scores. A decision tree plot model demonstrated that about 60% of the cases could be detected by the HRS-SR alone while the rest required a combination of HRS-SR and GAD-7 scores. ML evaluation metrics showed satisfactory performance, with a Matthews Correlation Coefficient of 55%; Area Under Curve (AUC), 79%; a Negative Predictive Value of 76%; and a False Negative Rate of 24%.ConclusionsStudy findings strongly suggest that ML can, in the future, play a significant role in the risk assessment of psychiatric disorders prior to face-to-face consultation. By using AI to scan big data questionnaire responses, wait time for seriously ill patients can be substantially cut, and prognoses substantially improved.

Performance Evaluation on E-Commerce Recommender System based on KNN, SVD, CoClustering and Ensemble Approaches

E-commerce recommender systems (RS) nowadays are essential for promoting products. These systems are expected to offer personalized recommendations for users based on the user preference. This can be achieved by employing cutting-edge technology such as artificial intelligence (AI) and machine learning (ML). Tailored recommendations for users can boost user experience in using the application and hence increase income as well as the reputation of a company. The purpose of this study is to investigate popular ML methods for e-commerce recommendation and study the potential of ensemble methods to combine the strengths of individual approaches. These recommendations are derived from a multitude of factors, including users' prior purchases, browsing history, demographic information, and others. To forecast the interests and preferences of users, several techniques are chosen to be investigated in this study, which include Singular Value Decomposition (SVD), k-Nearest Neighbor Baseline (KNN Baseline) and CoClustering. In addition, several evaluation metrics including the fraction of concordant pairs (FCP), mean absolute error (MAE), root mean square error (RMSE) and normalized discounted cumulative gain (NDCG) will be used to assess how well different techniques work. To provide a better understanding, the outcomes produced in this study will be incorporated into a graphical user interface (GUI).

A Comprehensive Survey on Deep Clustering: Taxonomy, Challenges, and Future Directions

Clustering is a fundamental machine learning task which aims at assigning instances into groups so that similar samples belong to the same cluster while dissimilar samples belong to different clusters. Shallow clustering methods usually assume that data are collected and expressed as feature vectors within which clustering is performed. However, clustering high-dimensional data, such as images, texts, videos, and graphs, poses significant challenges for clustering tasks, such as indiscriminate representation and intricate relationships among instances. Over the past decades, deep learning has achieved remarkable success in effective representation learning and modeling complex relationships. Motivated by these advancements, Deep Clustering seeks to improve clustering outcomes through deep learning techniques, garnering considerable interest from both academia and industry. Despite many contributions to this vibrant area of research, the lack of systematic analysis and a comprehensive taxonomy has hindered progress in this field. In this survey, we first explore how deep learning can be integrated into deep clustering and identify two fundamental components: the representation learning module and the clustering module. Then we summarize and analyze the representative design of these two modules. Furthermore, we introduce a novel taxonomy of deep clustering based on how these two modules interact, specifically through multistage, generative, iterative, and simultaneous approaches. In addition, we present well-known benchmark datasets, evaluation metrics, and open-source tools to clearly demonstrate different experimental approaches. Finally, we examine the practical applications of deep clustering and propose challenging areas for future research.

Performance Evaluation on E-Commerce Recommender System based on KNN, SVD, CoClustering and Ensemble Approaches

E-commerce recommender systems (RS) nowadays are essential for promoting products. These systems are expected to offer personalized recommendations for users based on the user preference. This can be achieved by employing cutting-edge technology such as artificial intelligence (AI) and machine learning (ML). Tailored recommendations for users can boost user experience in using the application and hence increase income as well as the reputation of a company. The purpose of this study is to investigate popular ML methods for e-commerce recommendation and study the potential of ensemble methods to combine the strengths of individual approaches. These recommendations are derived from a multitude of factors, including users' prior purchases, browsing history, demographic information, and others. To forecast the interests and preferences of users, several techniques are chosen to be investigated in this study, which include Singular Value Decomposition (SVD), k-Nearest Neighbor Baseline (KNN Baseline) and CoClustering. In addition, several evaluation metrics including the fraction of concordant pairs (FCP), mean absolute error (MAE), root mean square error (RMSE) and normalized discounted cumulative gain (NDCG) will be used to assess how well different techniques work. To provide a better understanding, the outcomes produced in this study will be incorporated into a graphical user interface (GUI).

Predicting low-velocity impact on the CNTs-reinforced plate using machine learning models

Purpose The study aims to predict a low-velocity impact on a plate reinforced with carbon nanotubes (CNTs) using machine learning models. Design/methodology/approach The first-order shear deformation plate theory (FSDT) is used to express the plate displacements filed. The Hertz nonlinear contact law is used to predict the contact between impactor and plate. Using the energy method and Hamilton’s principle, the motion equations are extracted. The six main parameters considered as inputs to machine learning models are CNTs percentage, impactor radius, plate thickness, plate length and width, CNTs distribution profile and impactor initial velocity. These input parameters are used to predict two impact targets including contact force and contact time. Findings As the values of the targets are continuous, the machine learning task is considered a regression problem. Therefore, this study uses different regression models to predict the targets. These regression models include linear regression, stochastic gradient descent regressor, Bayesian regression, partial least squares regression, Gaussian process regression, multilayer perceptron regressor, support vector regression and decision tree regression. To validate the effectiveness of the regression models, experiments are designed based on different evaluation metrics. The results of the experiments demonstrate that the machine learning models achieve promising performance in predicting the contact force and contact time based on the input parameters. Originality/value Due to the volume of high numerical calculations of impact mechanics to reach the response, the targets of the impact problem are predicted using a variety of machine learning methods.

Cyber-Securing Medical Devices Using Machine Learning: A Case Study of Pacemaker

This study aims to enhance the cybersecurity framework of pacemaker devices by identifying vulnerabilities and recommending effective strategies. The objectives are to pinpoint cybersecurity weaknesses, utilize machine learning to predict security breaches, and propose countermeasures based on analytical trends. The literature review highlights the transformation of pacemaker technology from basic, fixed-rate devices to sophisticated systems with wireless capabilities, which, while improving patient care, also introduce significant cybersecurity risks. These risks include unauthorized entry, data breaches, and life-threatening device malfunctions. The methodology in this study utilizes a quantitative research approach using the WUSTL-EHMS-2020 dataset, which includes network traffic features, patients' biometric features, and attack label. The step-by-step method of machine learning prediction includes data collection, data preprocessing, feature engineering, and models’ training using Support Vector Machines (SVM) and Gradient Boosting Machines (GBM). The implementation results used evaluation metrics like accuracy, precision, recall, and F1 score to show that GBM model outperformed the SVM model. The GBM model achieved higher accuracy of 95.1% compared to 92.5% for SVM, greater precision of 99.6% compared to 96.7% for SVM, better recall of 94.9% compared to 42.7% for SVM, and a higher F1 score of 76.3% compared to 59.0% for SVM, making GBM model more effective in predicting cybersecurity threats. This study concludes that GBM is an effective machine learning model for enhancing pacemaker cybersecurity by analyzing network traffic and biometric data patterns. Future recommendations for improving the pacemaker cybersecurity include implementing GBM model for threat predictions, integration with existing security measures, and regular model updates and retraining.

Sibling Discrimination Using Linear Fusion on Deep Learning Face Recognition Models

Facial recognition technology has revolutionised human identification, providing a non-invasive alternative to traditional biometric methods like signatures and voice recognition. The integration of deep learning has significantly enhanced the accuracy and adaptability of these systems, now widely used in criminal identification, access control, and security. Initial research focused on recognising full-frontal facial features, but recent advancements have tackled the challenge of identifying partially visible faces, a scenario that often reduces recognition accuracy. This study aims to identify siblings based on facial features, particularly in cases where only partial features like eyes, nose, or mouth are visible. Utilising advanced deep learning models such as VGG19, VGG16, VGGFace, and FaceNet, the research introduces a framework to differentiate between sibling images effectively. To boost discrimination accuracy, the framework employs a linear fusion technique that merges insights from all the models used. The methodology involves preprocessing image pairs, extracting embeddings with pre-trained models, and integrating information through linear fusion. Evaluation metrics, including confusion matrix analysis, assess the framework's robustness and precision. Custom datasets of cropped sibling facial areas form the experimental basis, testing the models under various conditions like different facial poses and cropped regions. Model selection emphasises accuracy and extensive training on large datasets to ensure reliable performance in distinguishing subtle facial differences. Experimental results show that combining multiple models' outputs using linear fusion improves the accuracy and realism of sibling discrimination based on facial features. Findings indicate a minimum accuracy of 96% across different facial regions. Although this is slightly lower than the accuracy achieved by a single model like VGG16 with full-frontal poses, the fusion approach provides a more realistic outcome by incorporating insights from all four models. This underscores the potential of advanced deep learning techniques in enhancing facial recognition systems for practical applications.

Hybrid RFSVM: Hybridization of SVM and Random Forest Models for Detection of Fake News

The creation and spreading of fake information can be carried out very easily through the internet community. This pervasive escalation of fake news and rumors has an extremely adverse effect on the nation and society. Detecting fake news on the social web is an emerging topic in research today. In this research, the authors review various characteristics of fake news and identify research gaps. In this research, the fake news dataset is modeled and tokenized by applying term frequency and inverse document frequency (TFIDF). Several machine-learning classification approaches are used to compute evaluation metrics. The authors proposed hybridizing SVMs and RF classification algorithms for improved accuracy, precision, recall, and F1-score. The authors also show the comparative analysis of different types of news categories using various machine-learning models and compare the performance of the hybrid RFSVM. Comparative studies of hybrid RFSVM with different algorithms such as Random Forest (RF), naïve Bayes (NB), SVMs, and XGBoost have shown better results of around 8% to 16% in terms of accuracy, precision, recall, and F1-score.

Baltic sea ice thickness estimation based on X-band SAR data and background information

Abstract In this study an operational sea ice thickness (SIT) estimation algorithm, based on HH-polarized X-band synthetic aperture radar (SAR) data, background information from the most recent, typically from the previous day, available daily Baltic Sea ice chart and the operational Finnish Meteorological Institute (FMI) thermodynamic ice model, was developed and evaluated. The algorithm was designed to complement the C-band SAR SIT algorithm developed earlier at FMI and applied daily as part of the operational Copernicus Marine Service (CMS). The X-band SIT algorithm was developed by utilizing the sea ice thickness measurements made onboard the Finnish and Swedish ice breakers during two winters seasons: 2021–2022 and 2022–2023. The former season measurements were used for defining the algorithm parameters and the later season for evaluation of the algorithm performance. According to the evaluation metrics the X-band algorithm performance is slightly better than that of the operational CMS C-band SAR SIT algorithm, indicating its suitability for operational use in CMS.

Deep spatial-temporal information fusion dynamic graph convolutional network for traffic flow prediction

Abstract Predicting traffic flow is vital for advancing intelligent transportation systems, but achieving precise forecasts remains challenging. To deeply explore the complex and diverse dynamic spatio-temporal correlation of traffic data, this study proposes a multi-information fusion spatio-temporal dynamic graph convolution model for traffic flow prediction. In the model construction, firstly, a dynamic graph generation module is designed to fully capture the complex spatial correlation of the traffic network and model the changing spatio-temporal interaction; secondly, the spatial self-attention mechanism is used to dynamically focus on the spatial relationship between different nodes, and combined with the dynamic graph convolution network to extract the spatial dynamic correlation features of the road network to study deep spatiotemporal information; Finally, existing models rarely consider other traffic information related to traffic flow, and a spatiotemporal information fusion module is designed to enhance the ability of the model to perceive information, thereby improving the performance of traffic flow prediction models. Experiments are conducted on four real traffic datasets using three performance evaluation metrics and 18 baseline models. The results show that the model has higher prediction accuracy.

The construction of international wetland urban ecological security pattern coupled with MSPA and ESF.

Creating an ecological security pattern is crucial for balancing the sustainable development of areas where human activities and the natural environment intersect. Using Hefei, an internationally recognized Wetland City, as a case study, we extracted ecological sources through ecological service function (ESF) analysis and morphological spatial pattern analysis (MSPA) core area connectivity analysis. Based on these ecological sources, we developed an ecological resistance surface system and identified ecological corridors and nodes using circuit theory. The findings are as follows: (1) Ecological source areas: The primary ecological sources in Hefei are located in the water bodies, forested areas, and scattered grasslands in the central and eastern parts of the city. This ecological source area covers 978.96 km2, which constitutes 8.55% of the city's total area. (2) Ecological corridors: Hefei contains 43 ecological corridors with a total length of 940.3km, averaging 21.87km each. These corridors are crucial for maintaining ecological connectivity and facilitating species movement. (3) Ecological nodes: There are 13 significant ecological nodes in Hefei, including 6 ecological obstacle points and 7 ecological pinpoints. These nodes play a vital role in supporting ecological processes and ensuring habitat connectivity. (4) Evaluation metrics: The α and β values for the source identification method that integrates MSPA with ecological service functions were 2.15 and 0.8, respectively, which are higher than those of the control group. Conversely, the γ-value was 0.18, lower than that of the control group. These results indicate that the combined ecological source extraction method provides significant advantages in terms of ecological corridor integrity, connectivity, and ecological flow management.

The Distinction Between Search and Discovery Systems : A Scholarly Analysis

This comprehensive analysis explores the fundamental distinctions between search and discovery systems in the context of information retrieval and user interaction. The article examines the underlying principles, methodologies, and user experiences associated with each system, elucidating their unique roles in the modern digital landscape. By comparing key aspects such as user intent, interaction models, algorithmic foundations, and evaluation metrics, the article highlights the complementary strengths of these paradigms. The research also investigates emerging hybrid approaches that combine elements of both search and discovery, addressing evolving user expectations and the challenges of information overload. Through this in-depth exploration, the article contributes to the ongoing discourse on the future of information access and informs the development of more effective, user-centric systems capable of meeting diverse information needs in an increasingly complex digital ecosystem.

An efficient interpretable stacking ensemble model for lung cancer prognosis

Lung cancer significantly contributes to global cancer mortality, posing challenges in clinical management. Early detection and accurate prognosis are crucial for improving patient outcomes. This study develops an interpretable stacking ensemble model (SEM) for lung cancer prognosis prediction and identifies key risk factors. Using a Kaggle dataset of 1000 patients with 22 variables, the model classifies prognosis into Low, Medium, and High-risk categories. The bootstrap method was employed for evaluation metrics, while SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-agnostic Explanations) assessed model interpretability. Results showed SEM's superior interpretability over traditional models, such as Random Forest, Logistic Regression, Decision Tree, Gradient Boosting Machine, Extreme Gradient Boosting Machine, and Light Gradient Boosting Machine. SEM achieved an accuracy of 98.90 %, precision of 98.70 %, F1 score of 98.85 %, sensitivity of 98.77 %, specificity of 95.45 %, Cohen’s kappa value of 94.56 %, and an AUC of 98.10 %. The SEM demonstrated robust performance in lung cancer prognosis, revealing chronic lung cancer and genetic risk as major factors.

Characterizing Rock-Breaking Performance of PDC Cutters via Stability Metrics and Energy consumption in FDEM Simulations

As oil production increasingly transitions from shallow to deep formation, the need for efficient PDC cutters to drill through deep hard rock becomes paramount. Currently, there is a lack of effective methods for optimizing cutter shapes and their cutting parameters. This study addresses this gap by developing a rock-breaking mechanism model for PDC cutters and proposing a Stability Index (SI) based on cutting force variations during the cutting process. Using the FDEM method, we simulated the rock-breaking processes of planar, stinger, and machete cutters under various cutting depths and angles. SI and Mechanical Specific Energy (MSE) were employed as evaluation metrics to analyze the rock-breaking characteristics and determine the optimal cutting parameters for each cutter type. The results indicate that at shallow cutting depths, cutting force changes are minimal, predominantly causing plastic damage to the rock, resulting in lower stress and reduced wear on the PDC cutters. Conversely, at greater cutting depths, significant fluctuations in cutting force occur, leading to brittle failure and the generation of large cuttings. This sudden increase in cutting force can accelerate rock breaking but also increases the risk of cutter wear. Our findings suggest that the planar cutter offers stable performance with high cutting force requirements, optimal at a cutting depth of 1.5 mm and an angle of less than 20°. The stinger cutter, requiring lower cutting forces, performs best at a cutting depth of 2 mm and an angle of 20°. The machete cutter exhibits characteristics similar to the stinger cutter at shallow depths and the planar cutter at deeper depths, making it suitable for drilling at a depth of 2 mm and an angle of less than 10°. By balancing SI and MSE, this study provides a comprehensive approach to optimizing PDC cutter performance, enhancing drilling efficiency while minimizing cutter wear.

Predicting stroke occurrences: a stacked machine learning approach with feature selection and data preprocessing

Stroke prediction remains a critical area of research in healthcare, aiming to enhance early intervention and patient care strategies. This study investigates the efficacy of machine learning techniques, particularly principal component analysis (PCA) and a stacking ensemble method, for predicting stroke occurrences based on demographic, clinical, and lifestyle factors. We systematically varied PCA components and implemented a stacking model comprising random forest, decision tree, and K-nearest neighbors (KNN).Our findings demonstrate that setting PCA components to 16 optimally enhanced predictive accuracy, achieving a remarkable 98.6% accuracy in stroke prediction. Evaluation metrics underscored the robustness of our approach in handling class imbalance and improving model performance, also comparative analyses against traditional machine learning algorithms such as SVM, logistic regression, and Naive Bayes highlighted the superiority of our proposed method.

Evaluating Expert-Layperson Agreement in Identifying Jargon Terms in Electronic Health Record Notes: Observational Study.

Studies have shown that patients have difficulty understanding medical jargon in electronic health record (EHR) notes, particularly patients with low health literacy. In creating the NoteAid dictionary of medical jargon for patients, a panel of medical experts selected terms they perceived as needing definitions for patients. This study aims to determine whether experts and laypeople agree on what constitutes medical jargon. Using an observational study design, we compared the ability of medical experts and laypeople to identify medical jargon in EHR notes. The laypeople were recruited from Amazon Mechanical Turk. Participants were shown 20 sentences from EHR notes, which contained 325 potential jargon terms as identified by the medical experts. We collected demographic information about the laypeople's age, sex, race or ethnicity, education, native language, and health literacy. Health literacy was measured with the Single Item Literacy Screener. Our evaluation metrics were the proportion of terms rated as jargon, sensitivity, specificity, Fleiss κ for agreement among medical experts and among laypeople, and the Kendall rank correlation statistic between the medical experts and laypeople. We performed subgroup analyses by layperson characteristics. We fit a beta regression model with a logit link to examine the association between layperson characteristics and whether a term was classified as jargon. The average proportion of terms identified as jargon by the medical experts was 59% (1150/1950, 95% CI 56.1%-61.8%), and the average proportion of terms identified as jargon by the laypeople overall was 25.6% (22,480/87,750, 95% CI 25%-26.2%). There was good agreement among medical experts (Fleiss κ=0.781, 95% CI 0.753-0.809) and fair agreement among laypeople (Fleiss κ=0.590, 95% CI 0.589-0.591). The beta regression model had a pseudo-R2 of 0.071, indicating that demographic characteristics explained very little of the variability in the proportion of terms identified as jargon by laypeople. Using laypeople's identification of jargon as the gold standard, the medical experts had high sensitivity (91.7%, 95% CI 90.1%-93.3%) and specificity (88.2%, 95% CI 86%-90.5%) in identifying jargon terms. To ensure coverage of possible jargon terms, the medical experts were loose in selecting terms for inclusion. Fair agreement among laypersons shows that this is needed, as there is a variety of opinions among laypersons about what is considered jargon. We showed that medical experts could accurately identify jargon terms for annotation that would be useful for laypeople.

A Method Based on Deep Learning for Severe Convective Weather Forecast: CNN-BiLSTM-AM (Version 1.0)

In this study, we propose a model called CNN-BiLSTM-AM that utilizes deep learning techniques to forecast severe convective weather based on ERA5 hourly data and observations. The model integrates a CNN with a Bidirectional Long Short-Term Memory (BiLSTM) system and an Attention Mechanism (AM). The CNN is tasked with extracting features from the input data, while the BiLSTM effectively captures temporal dependencies. The AM enhances the results by considering the impact of past feature states on severe weather phenomena. Additionally, we assess the performance of our model in comparison to traditional network architectures, including ConvLSTM, Predrnn++, CNN, FC-LSTM, and LSTM. Our results indicate that the CNN-BiLSTM-AM model exhibits superior accuracy in precipitation forecasting. Especially with the extension of the forecast time, the model performs well across multiple evaluation metrics. Furthermore, an interpretability analysis of the convective weather mechanisms utilizing machine learning highlights the critical role of total precipitable water (PWAT) in short-term heavy precipitation forecasts. It also emphasizes the significant impact of regional variables on convective weather patterns and the role of convective available potential energy (CAPE) in fostering conditions conducive to convection. These findings not only confirm the effectiveness of deep learning in the automatic identification of severe weather features but also validate the suitability of the sample dataset employed. Given its remarkable performance and robustness, we advocate for the adoption of this model to enhance the forecast of severe convective weather across various business applications.