Improve Model Performance Research Articles

A Self-Supervised Learning Based Framework for Eyelid Malignant Melanoma Diagnosis in Whole Slide Images.

Eyelid malignant melanoma (MM) is a rare disease with high mortality. Accurate diagnosis of such disease is important but challenging. In clinical practice, the diagnosis of MM is currently performed manually by pathologists, which is subjective and biased. Since the heavy manual annotation workload, most pathological whole slide image (WSI) datasets are only partially labeled (without region annotations), which cannot be directly used in supervised deep learning. For these reasons, it is of great practical significance to design a laborsaving and high data utilization diagnosis method. In this paper, a self-supervised learning (SSL) based framework for automatically detecting eyelid MM is proposed. The framework consists of a self-supervised model for detecting MM areas at the patch-level and a second model for classifying lesion types at the slide level. A squeeze-excitation (SE) attention structure and a feature-projection (FP) structure are integrated to boost learning on details of pathological images and improve model performance. In addition, this framework also provides visual heatmaps with high quality and reliability to highlight the likely areas of the lesion to assist the evaluation and diagnosis of the eyelid MM. Extensive experimental results on different datasets show that our proposed method outperforms other state-of-the-art SSL and fully supervised methods at both patch and slide levels when only a subset of WSIs are annotated. It should be noted that our method is even comparable to supervised methods when all WSIs are fully annotated. To the best of our knowledge, our work is the first SSL method for automatic diagnosis of MM at the eyelid and has a great potential impact on reducing the workload of human annotations in clinical practice.

Hybrid solar irradiance nowcasting and forecasting with the SCOPE method and convolutional neural networks

We use a full year (2018) of GOES-R satellite data to produce 5-minute resolved information on cloud coverage for 7 Surface Radiation Budget Network (SURFRAD). The remote sensing images are then processed using the Spectral Cloud Optical Property Estimation (SCOPE) method in conjunction with Convolution Neural Networks (CNNs) for nowcasts and 1-hour ahead forecasts. We propose and compare two CNN based models for GHI now-casting with a 1-hour forecast horizon and a time resolution that processes data every 5 minutes: one enhanced with the SCOPE method for cloud estimation and another without it. The inclusion of SCOPE-derived information significantly improves model performance, yielding an average RMSE of 44.7 versus 68.9 W/m2 for the model without SCOPE information. The present work underscores the efficacy of even basic CNNs in interpreting satellite imagery combined with atmospheric models of the atmosphere to estimate ground irradiance accurately. The hybrid SCOPE-CNN model outperforms the basic CNN model that relies solely on 10 longwave channels, indicating the relevance of SCOPE’s physical features in enhancing predictions across various solar micro-climates. Further advancements using Convolutional Long Short-Time Memory (ConvLSTM) schemes lead to the development of a SCOPE-CNN-ConvLSTM model, showcasing significant enhancements over smart persistence across all conditions.

Whoosh User Sentiment Analysis on Social Media Using Word2Vec and the Best Naïve Bayes Probability Model

By using the Twitter microblogging feature, users can post short tweets with limited characters that express their thoughts and opinions regarding a matter. The newest transportation in Indonesia, a high-speed train namely Whoosh is one of the things that Twitter users responded to. This latest transportation has led to the emergence of opinions from the Indonesian people which are shared publicly in various media, one of which is social media. Therefore, to make it easier for business people or companies to understand public opinion regarding service improvements in the future, sentiment analysis on social media is needed to determine user opinions regarding high-speed train transportation. In this research, sentiment analysis of high-speed train users will be carried out on social media Twitter using Word2Vec and Naïve Bayes as classification methods. In this research, a comparison of Naïve Bayes models will also be carried out to find out the best Naïve Bayes method opportunity model. Simultaneously, the Word2vec feature extraction method was chosen because Word2Vec can be used to improve model performance and increase the accuracy of sentiment classification. This research found that the Word2Vec Skip-Gram model outperformed the Word2Vec CBOW model. The best model obtained was the use of the Gaussian Naïve Bayes and Word2Vec Skip-Gram models with an accuracy score of 77.18%, precision 70.35%, recall 76.09%, and f1-score 73.10%.

Censorship data-driven DNS resolution anomaly detection: An ensemble algorithm model with multivariate feature fusion

The Domain Name System (DNS), as a critical component of Internet infrastructure, is crucial for maintaining network stability and security. However, vulnerabilities in the DNS resolution process make it susceptible to various network attacks. Current DNS resolution anomaly detection methods are challenged by the scarcity of diverse anomalous sample data and the difficulty in accurately capturing such anomalies. To address those challenges, we firstly introduce a proactive anomaly detection method based on bidirectional national censorship behavior, abbreviated as CB-BiDAM. This method can collect diverse anomalous resolution data from multiple network spaces, covering common types of resolution anomalies and effectively solving the problem of sample scarcity. Further, we extract multidimensional features from four key aspects: response content, DNS attributes, resolution paths, and timing, to gain a deeper understanding of the relationship between multifaceted information in the DNS resolution process and anomalous events. Finally, based on these multidimensional features, we construct a DNS resolution anomaly detection model named DRADC, using a stacked ensemble approach. Through a series of comparative experiments, the DRADC model significantly outperforms existing machine learning algorithms in key metrics such as accuracy (97.48%), recall (96.87%), and F1 score (97.09%). Feature ablation experiments further demonstrates that incorporating multidimensional features significantly improves model performance, with a 3% increase in accuracy and a 3.5% increase in precision compared to models relying solely on response content features. By providing an accurate detection model for DNS resolution anomalies, this study aids network administrators in more effectively identifying and countering network attacks. Additionally, our research also contributes to the detection and response to domain censorship mechanisms, which is crucial for maintaining the openness and free flow of the Internet.

Improvement of drought forecasting by means of various machine learning algorithms and wavelet transformation

AbstractDrought, which is defined as a decrease in average rainfall amounts, is one of the most insidious natural disasters. When it starts, people may not be aware of it, which is why droughts are difficult to monitor. Scientists have long been working to predict and monitor droughts. For this purpose, they have developed many methods, such as drought indices, one of which is the Standardized Precipitation Index (SPI). In this study, SPI was used to detect droughts, and machine learning algorithms, including support vector machines (SVM), artificial neural networks, random forest, and decision tree, were used to predict droughts. In addition, 3 different statistical criteria, which are correlation coefficient (r), root mean square error (RMSE), and Nash–Sutcliffe efficiency (NSE), were used to investigate model performance values. The wavelet transform (WT) was also applied to improve model performance. One of the areas most impacted by droughts in Turkey is the Konya Closed Basin, which is geographically positioned in the center of the country and is among the top grain-producing regions in Turkey. The Apa Dam is one of the most significant water resources in the area. It provides water to many fertile fields in its vicinity and is affected by droughts which is why it was selected as a study area. Meteorological data, such as monthly precipitation, that could represent the region were obtained between 1955 and 2020 from the general directorate of state water works and the general directorate of meteorology. According to the findings, the M04 model, whose input structure was developed using SPI, various time steps, data delayed up to 5 months, and monthly precipitation data from the preceding month (time t − 1), produced the best results out of all the models examined using machine learning algorithms. Among machine learning algorithms, SVM has achieved the most successful results not only before applying WT but also after WT. The best results were obtained from M04, in which SVM with WT was used (NSE = 0.9942, RMSE = 0.0764, R = 0.9971).

Fairness gaps in Machine learning models for hospitalization and emergency department visit risk prediction in home healthcare patients with heart failure

ObjectivesThis study aims to evaluate the fairness performance metrics of Machine Learning (ML) models to predict hospitalization and emergency department (ED) visits in heart failure patients receiving home healthcare. We analyze biases, assess performance disparities, and propose solutions to improve model performance in diverse subpopulations. MethodsThe study used a dataset of 12,189 episodes of home healthcare collected between 2015 and 2017, including structured (e.g., standard assessment tool) and unstructured data (i.e., clinical notes). ML risk prediction models, including Light Gradient-boosting model (LightGBM) and AutoGluon, were developed using demographic information, vital signs, comorbidities, service utilization data, and the area deprivation index (ADI) associated with the patient’s home address. Fairness metrics, such as Equal Opportunity, Predictive Equality, Predictive Parity, and Statistical Parity, were calculated to evaluate model performance across subpopulations. ResultsOur study revealed significant disparities in model performance across diverse demographic subgroups. For example, the Hispanic, Male, High-ADI subgroup excelled in terms of Equal Opportunity with a metric value of 0.825, which was 28% higher than the lowest-performing Other, Female, Low-ADI subgroup, which scored 0.644. In Predictive Parity, the gap between the highest and lowest-performing groups was 29%, and in Statistical Parity, the gap reached 69%. In Predictive Equality, the difference was 45%. Discussion and ConclusionThe findings highlight substantial differences in fairness metrics across diverse patient subpopulations in ML risk prediction models for heart failure patients receiving home healthcare services. Ongoing monitoring and improvement of fairness metrics are essential to mitigate biases.

Scale Correction Cascade Smooth Loss in GANs to Improve Model Performance

Scale Correction Cascade Smooth Loss in GANs to Improve Model Performance

Towards Reliable Healthcare LLM Agents: A Case Study for Pilgrims during Hajj

There is a pressing need for healthcare conversational agents with domain-specific expertise to ensure the provision of accurate and reliable information tailored to specific medical contexts. Moreover, there is a notable gap in research ensuring the credibility and trustworthiness of the information provided by these healthcare agents, particularly in critical scenarios such as medical emergencies. Pilgrims come from diverse cultural and linguistic backgrounds, often facing difficulties in accessing medical advice and information. Establishing an AI-powered multilingual chatbot can bridge this gap by providing readily available medical guidance and support, contributing to the well-being and safety of pilgrims. In this paper, we present a comprehensive methodology aimed at enhancing the reliability and efficacy of healthcare conversational agents, with a specific focus on addressing the needs of Hajj pilgrims. Our approach leverages domain-specific fine-tuning techniques on a large language model, alongside synthetic data augmentation strategies, to optimize performance in delivering contextually relevant healthcare information by introducing the HajjHealthQA dataset. Additionally, we employ a retrieval-augmented generation (RAG) module as a crucial component to validate uncertain generated responses, which improves model performance by 5%. Moreover, we train a secondary AI agent on a well-known health fact-checking dataset and use it to validate medical information in the generated responses. Our approach significantly elevates the chatbot’s accuracy, demonstrating its adaptability to a wide range of pilgrim queries. We evaluate the chatbot’s performance using quantitative and qualitative metrics, highlighting its proficiency in generating accurate responses and achieve competitive results compared to state-of-the-art models, in addition to mitigating the risk of misinformation and providing users with trustworthy health information.

Dual hypergraphs with feature weighted and latent space learning for the diagnosis of Alzheimer’s disease

In recent years of research on the diagnosis of Alzheimer’s disease, capturing data relationships can help improve model performance. However, the simple graph structure can only capture pairwise relationships between data, which cannot model the complex data relationships in real situations. In addition, the redundant features and noise from the original data space also harm the model performance. Dual hypergraphs with feature-weighted and latent space learning (DHFWLSL) method for Alzheimer’s disease diagnosis are proposed to solve the abovementioned problems. Hypergraphs are used to capture higher-order data relationships and construct hypergraphs for both sample and feature relationships to fully utilize the auxiliary ability of data relationships for model classification. In order to remove noise and redundant features from the original data space, latent space learning is utilized to project the original data into the latent space, and then data is projected into the label space for diagnosis. Finally, a feature-weighted strategy helps to improve the model performance further. An alternating optimization algorithm with the proof of convergence is used to solve the proposed model. Sufficient experiments support the validity and advantages of the proposed model.

Temporal heterogeneity in the performance of machine learning models for PM2.5 concentration estimation

Machine learning (ML) methods have been applied extensively to simulate air pollutant concentrations and assess individual exposure in epidemiological studies. However, there is still a paucity of research on the temporal heterogeneity of ML model performance and the impact of dataset size. To explore the temporal heterogeneity in model performance when estimating daily concentrations of fine particulate matter (PM2.5) across China in 2021, we compared five decision tree-based ML models (Random Forest (RF), Categorical Boosting (CatBoost), Gradient Boost Regression Tree (GBRT), eXtreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM)) across daily scales within three distinct timeframes. The performance of all models was evaluated using cross-validation. We observed that the performance of ML models varied with time, which showed a significant correlation with PM2.5 concentration. Among the 365 days in 2021, RF model performed best, the annual mean R2 was 0.86, with a minimum of 0.84 and a maximum of up to 0.95. For RF, we chose a cubic polynomial curve to fit the relationship between model performance and PM2.5 concentrations, and based on this, we devised a model selection strategy for different time scales, achieving an accuracy rate of up to 79.45 %, with the selected models having an average R2 of 0.85, and a maximum of up to 0.95. Additionally, we found that increasing the dataset size did not significantly improve model performance. Instead, it resulted in considerably longer runtime and increased memory usage. The methodology and findings of this study hold significant value for advancing the development of more efficient and precise modeling approaches for air pollutant concentrations. Furthermore, this research provides a foundation for regional air pollutant governance and future health-related research.

Improving learning-based birdsong classification by utilizing combined audio augmentation strategies

In ecology, changes in environmental conditions are often closely linked to shifts in species diversity. This relationship can be investigated by analyzing avian vocalizations, which are robust indicators of trends in biodiversity. Within this contribution, we explored various data augmentation techniques and deep learning strategies for the classification of birdsong within natural soundscapes. For this purpose, we employed three fundamental deep neural network architectures, such as vision transformers, to classify 397 different bird species. To improve both the accuracy and generalizability of our models, we incorporated up to 19 well-established data augmentation techniques commonly used in audio classification. This included an iterative selection process where only augmentations that enhanced classification performance were selected. The primary augmentation technique involved the integration of various noise samples and non-bird audio elements, which significantly improved model performance as assessed on the BirdCLEF 2021 data set. Individual augmentations achieved F1-scores from 48.0 % (vertical flip) to 72.6 % (primary background noise soundscapes). Through the strategic combination of key techniques – namely simulated pink noise, interspecies sound mixing, and loudness normalization – we achieved a top F1-score of 73.7%. Depending on the selected classification model, this corresponds to an improvement by 4.81 % to 10.5 %. Improvements and deteriorations of all applied augmentation techniques appeared to be robust across our three evaluated models. Therefore, our approach highlights the potential of sophisticated audio augmentations in refining the accuracy and robustness of birdsong classification models.

Klasifikasi Jenis Jamur Edible Menggunakan Convolutional Neural Network: Studi Kasus pada Jamur Tiram, Enoki, dan Truffle

This study uses a Convolutional Neural Network (CNN) to develop a mushroom type classification model that can differentiate between truffles, enoki and oyster mushrooms very accurately. The dataset consisting of mushroom images is collected from various sources and processed through data augmentation techniques such as rotation, zoom, flip, and shifting to improve model generalization. For final classification, the CNN model used consists of several convolution and pooling layers, followed by a dense layer. The study results show that this model has a validation accuracy of 79.31% after twenty training epochs. Although there were some misclassifications between oyster and enoki mushrooms, data augmentation techniques were shown to improve model performance. This research shows that CNN can be used effectively to classify various types of fungi. This also shows that CNN has many potential applications in the agricultural and food industries. Increasing datasets, studying more complex model architectures, and using advanced data augmentation methods such as Generative Adversarial Networks (GANs) are some suggestions for further research.

An updated variable RBE model for proton therapy

Objective. While a constant relative biological effectiveness (RBE) of 1.1 forms the basis for clinical proton therapy, variable RBE models are increasingly being used in plan evaluation. However, there is substantial variation across RBE models, and several new in vitro datasets have not yet been included in the existing models. In this study, an updated in vitro proton RBE database was collected and used to examine current RBE model assumptions, and to propose an up-to-date RBE model as a tool for evaluating RBE effects in clinical settings. Approach. A proton database (471 data points) was collected from the literature, almost twice the size of the previously largest model database. Each data point included linear-quadratic model parameters and linear energy transfer (LET). Statistical analyses were performed to test the validity of commonly applied assumptions of phenomenological RBE models, and new model functions were proposed for RBEmax and RBEmin (RBE at the lower and upper dose limits). Previously published models were refitted to the database and compared to the new model in terms of model performance and RBE estimates. Main results. The statistical analysis indicated that the intercept of the RBEmax function should be a free fitting parameter and RBE estimates were clearly higher for models with free intercept. RBEmin increased with increasing LET, while a dependency of RBEmin on the reference radiation fractionation sensitivity ( α/βx ) did not significantly improve model performance. Evaluating the models, the new model gave overall lowest RMSE and highest R2 score. RBE estimates in the distal part of a spread-out-Bragg-peak in water ( α/βx = 2.1 Gy) were 1.24–1.51 for original models, 1.25–1.49 for refits and 1.42 for the new model. Significance. An updated RBE model based on the currently largest database among published phenomenological models was proposed. Overall, the new model showed better performance compared to refitted published RBE models.

A Microvascular Segmentation Network Based on Pyramidal Attention Mechanism.

The precise segmentation of retinal vasculature is crucial for the early screening of various eye diseases, such as diabetic retinopathy and hypertensive retinopathy. Given the complex and variable overall structure of retinal vessels and their delicate, minute local features, the accurate extraction of fine vessels and edge pixels remains a technical challenge in the current research. To enhance the ability to extract thin vessels, this paper incorporates a pyramid channel attention module into a U-shaped network. This allows for more effective capture of information at different levels and increased attention to vessel-related channels, thereby improving model performance. Simultaneously, to prevent overfitting, this paper optimizes the standard convolutional block in the U-Net with the pre-activated residual discard convolution block, thus improving the model's generalization ability. The model is evaluated on three benchmark retinal datasets: DRIVE, CHASE_DB1, and STARE. Experimental results demonstrate that, compared to the baseline model, the proposed model achieves improvements in sensitivity (Sen) scores of 7.12%, 9.65%, and 5.36% on these three datasets, respectively, proving its strong ability to extract fine vessels.

Classification of autonomous vehicle crash severity: Solving the problems of imbalanced datasets and small sample size

Only a few researchers have shown how environmental factors and road features relate to Autonomous Vehicle (AV) crash severity levels, and none have focused on the data limitation problems, such as small sample sizes, imbalanced datasets, and high dimensional features. To address these problems, we analyzed an AV crash dataset (2019 to 2021) from the California Department of Motor Vehicles (CA DMV), which included 266 collision reports (51 of those causing injuries). We included external environmental variables by collecting various points of interest (POIs) and roadway features from Open Street Map (OSM) and Data San Francisco (SF). Random Over-Sampling Examples (ROSE) and the Synthetic Minority Over-Sampling Technique (SMOTE) methods were used to balance the dataset and increase the sample size. These two balancing methods were used to expand the dataset and solve the small sample size problem simultaneously. Mutual information, random forest, and XGboost were utilized to address the high dimensional feature and the selection problem caused by including a variety of types of POIs as predictive variables. Because existing studies do not use consistent procedures, we compared the effectiveness of using the feature-selection preprocessing method as the first process to employing the data-balance technique as the first process.Our results showed that AV crash severity levels are related to vehicle manufacturers, vehicle damage level, collision type, vehicle movement, the parties involved in the crash, speed limit, and some types of POIs (areas near transportation, entertainment venues, public places, schools, and medical facilities). Both resampling methods and three data preprocessing methods improved model performance, and the model that used SMOTE and data-balancing first was the best. The results suggest that over-sampling and the feature selection method can improve model prediction performance and define new factors related to AV crash severity levels.

Hierarchy-Aware and Label Balanced Model for Hierarchical Text Classification

Hierarchical text classification, where labels can be modeled as a hierarchical structure, is a special multi-label text classification sub-task. Current methods mainly improve model performance by modeling label dependencies. The model HGCLR with excellent performance obtains hierarchy-aware text representations through contrastive learning, but the degree of hierarchical awareness is insufficient. To address this problem, a new multi-label negative supervision method is proposed to drive text representations of samples with more different labels farther away. In addition, to tackle the problem of label imbalance in hierarchical text classification, asymmetric loss is employed to compute the classification loss, so that the model focuses on learning from difficult samples and the contributions of positive and negative labels to the loss function tend to be balanced. In summary, based on HGCLR, by adding multi-label negative supervision and replacing the classification loss with asymmetric loss, we propose a hierarchy-aware and label balanced model (HALB) for hierarchical text classification. Experimental results demonstrate that HALB outperforms several classical models for hierarchical text classification. HALB achieves the best results on both Micro-F1 and Macro-F1 and obtains average improvements of 0.51% and 1.28% respectively on four datasets compared with the strongest baseline model HGCLR.

An efficient workflow for modelling high-dimensional spatial extremes

We develop a comprehensive methodological workflow for Bayesian modelling of high-dimensional spatial extremes that lets us describe both weakening extremal dependence at increasing levels and changes in the type of extremal dependence class as a function of the distance between locations. This is achieved with a latent Gaussian version of the spatial conditional extremes model that allows for computationally efficient inference with R-INLA. Inference is made more robust using a post hoc adjustment method that accounts for possible model misspecification. This added robustness makes it possible to extract more information from the available data during inference using a composite likelihood. The developed methodology is applied to the modelling of extreme hourly precipitation from high-resolution radar data in Norway. Inference is performed quickly, and the resulting model fit successfully captures the main trends in the extremal dependence structure of the data. The post hoc adjustment is found to further improve model performance.

IMG-12. CBTN BRAIN TUMOR SEGMENTATION INITIATIVE: UPDATES ON MODEL RELEASE, HGG SEGMENTATION, AND SURVIVAL ANALYSIS

Abstract BACKGROUND Assessment of treatment responses in pediatric brain tumors requires accurate tumor segmentation, particularly important for nonresectable tumors like diffuse midline glioma (DMG), including diffuse intrinsic pontine glioma (DIPGs). Evaluating tumor progression in these tumors relies on monitoring tumor size changes in longitudinal MRI exams which is challenged by their infiltrative growth patterns. Our team developed a comprehensive multi-institutional, multi-histology autosegmentation deep learning (DL) model leveraging data from the Children’s Brain Tumor Network (CBTN), involving continuous enhancements and enriched by an expanding dataset including post-surgical/post-treatment studies. METHODS Recently, we trained and validated this DL model (publicly available) on a multi-institutional dataset of 644 PBTs, including 273 pediatric-type diffuse high-grade gliomas (HGGs, including DMG/DIPGs), using pre-operative and post-operative/post-treatment multiparametric MRI scans. We tested the model in 23 longitudinal MRI exams of 8 DMGs from the PNOC003 and PNOC007 clinical trials and applied Cox regression for predicting overall survival (OS) using tumor volume, computed with our autosegmentation model, as a time-varying variable in 18 patients. RESULTS For HGGs in treatment-naïve sessions, our model demonstrated strong performance with median Dice scores of 0.89 for whole tumor and tumor core (encompassing all regions, except for edema), and 0.73 for enhancing tumor segmentation. In longitudinal MRI exams of DMGs from the aforementioned clinical trials, the model achieved median Dice scores of 0.79 for whole tumor, and 0.78 for enhancing tumor and tumor core segmentation. Cox regression showed the significance of longitudinal volumetric measurements (p=0.08) for predicting OS in DMGs. CONCLUSIONS Our approach provides reliable tumor segmentation in HGGs across diverse datasets, crucial for treatment response assessment. Looking ahead, we plan to validate our model on a larger set of longitudinal exams and use volumetric measurements to predict tumor progression. This ongoing effort aims to improve model performance, extend its generalizability, and mitigate potential model decay.

Implementation and Evaluation of Spatial Attention Mechanism in Apricot Disease Detection Using Adaptive Sampling Latent Variable Network.

In this study, an advanced method for apricot tree disease detection is proposed that integrates deep learning technologies with various data augmentation strategies to significantly enhance the accuracy and efficiency of disease detection. A comprehensive framework based on the adaptive sampling latent variable network (ASLVN) and the spatial state attention mechanism was developed with the aim of enhancing the model's capability to capture characteristics of apricot tree diseases while ensuring its applicability on edge devices through model lightweighting techniques. Experimental results demonstrated significant improvements in precision, recall, accuracy, and mean average precision (mAP). Specifically, precision was 0.92, recall was 0.89, accuracy was 0.90, and mAP was 0.91, surpassing traditional models such as YOLOv5, YOLOv8, RetinaNet, EfficientDet, and DEtection TRansformer (DETR). Furthermore, through ablation studies, the critical roles of ASLVN and the spatial state attention mechanism in enhancing detection performance were validated. These experiments not only showcased the contributions of each component for improving model performance but also highlighted the method's capability to address the challenges of apricot tree disease detection in complex environments. Eight types of apricot tree diseases were detected, including Powdery Mildew and Brown Rot, representing a technological breakthrough. The findings provide robust technical support for disease management in actual agricultural production and offer broad application prospects.

Application of Fashion Element Trend Prediction Model Integrating AM and EfficientNet-b7 Models in Art Design

With the rapid development of the fashion industry and the increasing diversification of consumer needs, accurately predicting the fashion trends of clothing elements has become an urgent problem in the field of art design. In order to solve these problems, this paper uses deep learning technology for fashion trend prediction and optimization of prediction models. Firstly, the EfficientNet-b7 model is constructed as an attribute predictor to accurately extract the attributes of clothing image elements. Then, based on user information, popular elements are grouped and counted to solve the problem of different populations having different opinions on popular trends. The prediction model is constructed based on the bidirectional long short-term memory network encoder decoder framework, which trains trend information as a whole and utilizes element coexistence relationships to assist in trend prediction. Meanwhile, the study uses random sampling method for clothing original adoption, and the experimental results show that the model considering coexistence relationship performs the best in terms of mean absolute error and mean absolute percentage error, which are 0.0132 and 14.68%, respectively. In addition, the model that introduces attention mechanism based on trend similarity has improved by 5.09% and 4.54% on two indicators compared to the latest model. The experimental results indicate that coexistence relationships can help improve the performance of prediction models. The attention mechanism based on trend similarity can further improve model performance by similarity comparison between historical information and changing trend, and selecting similar historical information as an important influencing factor for future trends.