Average Accuracy Value Research Articles

Weighted Combination of Łukasiewicz implication and Fuzzy Jaccard similarity in Hybrid Ensemble Framework (WCLFJHEF) for Gene Selection

A framework is developed for gene expression analysis by introducing fuzzy Jaccard similarity (FJS) and combining Łukasiewicz implication with it through weights in hybrid ensemble framework (WCLFJHEF) for gene selection in cancer. The method is called weighted combination of Łukasiewicz implication and fuzzy Jaccard similarity in hybrid ensemble framework (WCLFJHEF). While the fuzziness in Jaccard similarity is incorporated by using the existing Gödel fuzzy logic, the weights are obtained by maximizing the average F-score of selected genes in classifying the cancer patients. The patients are first divided into different clusters, based on the number of patient groups, using average linkage agglomerative clustering and a new score, called WCLFJ (weighted combination of Łukasiewicz implication and fuzzy Jaccard similarity). The genes are then selected from each cluster separately using filter based Relief-F and wrapper based SVMRFE (Support Vector Machine with Recursive Feature Elimination). A gene (feature) pool is created by considering the union of selected features for all the clusters. A set of informative genes is selected from the pool using sequential backward floating search (SBFS) algorithm. Patients are then classified using Naïve Bayes’(NB) and Support Vector Machine (SVM) separately, using the selected genes and the related F-scores are calculated. The weights in WCLFJ are then updated iteratively to maximize the average F-score obtained from the results of the classifier. The effectiveness of WCLFJHEF is demonstrated on six gene expression datasets. The average values of accuracy, F-score, recall, precision and MCC over all the datasets, are 95%, 94%, 94%, 94%, and 90%, respectively. The explainability of the selected genes is shown using SHapley Additive exPlanations (SHAP) values and this information is further used to rank them. The relevance of the selected gene set are biologically validated using the KEGG Pathway, Gene Ontology (GO), and existing literatures. It is seen that the genes that are selected by WCLFJHEF are candidates for genomic alterations in the various cancer types. The source code of WCLFJHEF is available at http://www.isical.ac.in/~shubhra/WCLFJHEF.html.

An artificial intelligence approach to predict infants’ health status at birth

BackgroundMachine learning could be used for prognosis/diagnosis of maternal and neonates’ diseases by analyzing the data sets and profiles obtained from a pregnant mother. PurposeWe aimed to develop a prediction model based on machine learning algorithms to determine important maternal characteristics and neonates’ anthropometric profiles as the predictors of neonates’ health status. MethodsThis study was conducted among 1280 pregnant women referred to healthcare centers to receive antenatal care. We evaluated several machine learning methods, including support vector machine (SVM), Ensemble, K-Nearest Neighbor (KNN), Naïve Bayes (NB), and Decision tree classifiers, to predict newborn health state. ResultsThe minimum redundancy-maximum relevance (MRMR) algorithm revealed that variables, including head circumference of neonates, pregnancy intention, and drug consumption history during pregnancy, were top-scored features for classifying normal and unhealthy infants. Among the different classification methods, the SVM classifier had the best performance. The average values of accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC) in the test group were 75%, 75%, 76%, 76%, and 65%, respectively, for SVM model. ConclusionMachine learning methods can efficiently forecast the neonate's health status among pregnant women. This study proposed a new approach toward the integration of maternal data and neonate profiles to facilitate the prediction of neonates’ health status.

Research on the Innovation and Effectiveness of the Teaching Mode of Civics and Political Science Courses Supported by Mobile Computing Platforms

Abstract In today’s world, with the rapid development of information technology and scientific education, data technology has triggered important changes in ideological and political education in colleges and universities, effectively promoting the transformation of educational concepts, ways of thinking, educational methods, and practice modes. Based on the research background under big data, this study investigates the traditional knowledge-tracking model, introduces the attention mechanism and the characteristics of ideological and political learners on the basis of the traditional model, and proposes an improved knowledge-tracking model. At the same time, a mobile teaching platform was constructed, then the improved knowledge tracking model was tested, and the effect of the knowledge platform application on the relevant dataset was explored. The results show that using the optimized knowledge state tracking model, the teacher’s teaching time can be adjusted according to the students’ state, and there is a good effect on the prediction of students’ grades, and the average grade of the students is only 2-6 points of error from the predicted grades. In terms of model application accuracy, the average value of low accuracy is 6.92%, the average value of medium accuracy is 12.9%, and the average value of high accuracy is 79.79%, so the model assesses students’ knowledge status with high accuracy. Finally, the optimized model was used to analyze the learning effect of the students and to conduct a survey on the effect of using satisfaction, and the survey concluded that the students were more satisfied with the model.

Classification of presence of missing teeth in each quadrant using deep learning artificial intelligence on panoramic radiographs of pediatric patients.

Early tooth loss in pediatric patients can lead to various complications, making quick and accurate diagnosis essential. This study aimed to develop a novel deep learning model for classification of missing teeth on panoramic radiographs in pediatric patients and to assess the accuracy. The study included patients aged 8-16 years who visited the Pusan National University Dental Hospital and underwent panoramic radiography. A total of 806 panoramic radiographs were retrospectively analyzed to determine the presence or absence of missing teeth for each tooth number. Moreover, each panoramic radiograph was divided into four quadrants, each of a smaller size, containing both primary and permanent teeth, generating 3224 data. Quadrants with missing teeth (n = 1457) were set as the experimental group, and quadrants without missing teeth (n = 1767) were set as the control group. The data were split into training and validation sets in a 4:1 ratio, and a 5-fold cross-validation was conducted. A gradient-weighted class activation map was used to visualize the deep learning model. The average values of sensitivity, specificity, accuracy, precision, recall and F1-score of this deep learning model were 0.635, 0.814, 0.738, 0.730, 0.732 and 0.731, respectively. In the experimental group, the accuracy was the highest for missing canines and premolars, and the lowest for molars. The deep learning model exhibited a moderate to good distinguishing power with a classification performance of 0.730. This deep learning model and the newly defined small sized region of interest proved adequate for classifying the presence of missing teeth.

Influence of Artificial Intelligence on Forecasting Net Asset Value and Return Volatility in Indian Mutual Fund

Predicting the critical net asset value (NAV) in the financial market is difficult for investors and fund agencies. The present study introduces machine learning (ML) and deep learning (DL) models such as linear regression, deep long short-term memory recurrent neural network (DLSTM-RNN), and autoregressive integrated moving averages (ARIMA) for predicting the NAV. The five different equity sectoral technology mutual fund direct growth plans from January 2013 to December 2022 have been collected. The novelty of the current study is deeply examining, which ML or DL model devotedly predicts the NAV closing price. The major key findings of the experimental results proved that the DLSTM-RNN model makes statistically viable predictions, whereby the mean absolute percent error (MAPE) average prediction accuracy value is 0.02. Based on the accuracy of a superior model, we compute the annualized return volatility to compare the risk of investments with annual return periods over different time horizons. The Jarque-Bera statistics of the return volatility over time Gaussian distribution is rejected at the 0.01 level. Statistical paired t-test and Pearson correlation coefficient are used to compare the effects of the proposed three models. In addition, the benchmark portfolio strategy yields a Sharpe ratio of 7.0193 and the maximum drawdown is 0.3743. The AI performed deep LSTM neural network model simulation, especially when using a daily and monthly MAPE strategy giving 81% and 84% highest NAV prediction consistency than the linear regression and ARIMA models.

Optimasi Penerapan Algoritma Yolo dan Data Augmentasi dalam Klasifikasi Pakaian Tradisional Kebaya

Kebaya is one of Indonesia's traditional clothes which has a long history and is rich in culture. Each region or tribe in Indonesia has a kebaya with its own characteristics, including unique patterns, colors and decorations .Indonesian society want to know the kebaya from their area of origin but have difficulty distinguishing the kebayas because there are only slight differences in certain parts of the kebaya which are characteristic of kebaya from one area with kebaya from other regions. Therefore we need a system that can detect or classify traditional kebaya clothes in Indonesia. Researchers use the YOLOv8 version in making this system. The YOLOv8 algorithm goes through a classification system for kebaya clothes, processes such as annotation, preprocessing, augmentation, training to testing. The research results show good results with a confidence value of 90% - 97%. The final results of testing on 12 images show the output by knowing the kebaya class and good accuracy and producing an average accuracy value of 94%.

Aspect-Based Sentiment Analysis of Borobudur Temple Reviews Use Support Vector Machine Algorithm

As one of the most popular tourist attractions in Indonesia, Borobudur Temple is currently included in the top ten list of tourism priorities by the Ministry of Tourism. To increase the number of tourists, it is very important to maintain the quality of tourist attractions. Tourist growth is directly related to the number of online reviews of tourist attractions. Tourism managers need more than just reviewing good and negative sentiments to maintain and improve the quality of tourist attractions. Many aspects serve as benchmarks for visitors to come to a tourist spot, such as aspects of ticket prices, location, attractiveness, facilities, accessibility, visual image, and human resources. Therefore, sentiment analysis is needed for each of these aspects to find out aspects that need to be improved in order to increase the number of visitors. Support Vector Machine (SVM) is an algorithm used to categorize aspect-based sentiments. analyzed using SVM, the dataset must first be cleaned and normalized through pre-processing. The results of the analysis show that the aspects of accessibility and visual image need to be improved to maintain and increase the number of visitors. This is because these two aspects have the most negative reviews compared to other aspects. The results of model testing only get an average accuracy value of 0.8148 because the distribution of data for all aspects and reviews is not balanced.

Ship imaging trajectory extraction via an aggregated you only look once (YOLO) model

Maritime traffic community has paid a huge amount of focuses to establish maritime intelligent transportation infrastructure for the purpose of enhancing maritime traffic safety and efficiency. Maritime surveillance video is considered as a type of fundamental data sources for establishing intelligent maritime transportation infrastructure towards smart ship era. To that end, the study proposes an aggregated deep learning model-supported ship imaging trajectory extraction framework. The proposed framework starts by detecting ships from maritime images via a novel You Only Look Once (YOLO) model. More specifically, the proposed ship trajectory extraction framework obtains ship positions in a frame-by-frame manner via the proposed poly-YOLO module. Then, the proposed model maps ship positions in neighboring consecutive maritime images via an Enhanced Deep Sort (EDS) module. Experimental results suggest that the proposed ship trajectory extraction model achieves satisfactory performance due to that the average values of index multiple-object tracking accuracy (MOTa), recall rate (Rid) and index aggregated detection accuracy (Aggid) are larger than 89% (which outperform the comparison algorithms). The study can help varied maritime traffic participants obtain accurate on-site traffic situations in the smart ship era.

DEEP LEARNING ARCHITECTURE BASED ON CONVOLUTIONAL NEURAL NETWORK (CNN) IN IMAGE CLASSIFICATION

In current technological developments, Deep Learning is one of the most popular studies today, especially in the fields of machine learning and computer vision, GPU Acceration Technology is one of the reasons for the development of Deep Learning. Deep Learning has a very good ability to solve classic problems in the field of computer vision, one of which is in the case of object classification in images. one of the deep learning methods that is often used in image processing is the Convolution Neural Network (CNN) which is a development of the Multi Layer Perceptron method. This study uses the CNN architecture which consists of a convolution layer, as well as a fully connected layer, and will also determine the appropriate Optimizer and Loss function for CNN. The implementation of this method uses Google Colab (Tensorflow and Keras) with the Python programming language. In the training process using CNN, setting the number of epochs is done to improve accuracy in image classification, in the first scenario using epoch 20 produces an average accuracy of 99.45 with a loss value of 1.66. In the second scenario using epoch 15 produces an average accuracy value of 99.00 with a loss value of 2.92. then in the third scenario with a number of epochs 10 it produces an average accuracy value of 95.55 with a loss value of 95.55, while in the last scenario with a number of epochs 5 it produces an average accuracy value of 73.6 with a loss value of 51.92. From the 4 trial scenarios using the CNN method gives effective results and produces a fairly good accuracy value with an average accuracy and loss value of 99.99%. As well as the results of an average loss of 4.

Genomic predictions and GWAS for heat tolerance in pigs based on reaction norm models with performance records and data from public weather stations considering alternative temperature thresholds.

Genetic improvement of livestock productivity has resulted in greater production of metabolic heat and potentially greater susceptibility to heat stress. Various studies have demonstrated that there is genetic variability for heat tolerance and genetic selection for more heat tolerant individuals is possible. The rate of genetic progress tends to be greater when genomic information is incorporated into the analyses as more accurate breeding values can be obtained for young individuals. Therefore, this study aimed (1) to evaluate the predictive ability of genomic breeding values for heat tolerance based on routinely recorded traits, and (2) to investigate the genetic background of heat tolerance based on single-step genome-wide association studies for economically important traits related to body composition, growth and reproduction in Large White pigs. Pedigree information was available for 265,943 animals and genotypes for 8686 animals. The studied traits included ultrasound backfat thickness (BFT), ultrasound muscle depth (MDP), piglet weaning weight (WW), off-test weight (OTW), interval between farrowing (IBF), total number of piglets born (TNB), number of piglets born alive (NBA), number of piglets born dead (NBD), number of piglets weaned (WN) and weaning-to-estrus interval (IWE). The number of phenotypic records ranged from 6059 (WN) to 172,984 (TNB). Single-step genomic reaction norm predictions were used to calculate the genomic estimated breeding values for each individual. Predictions of breeding values for the validation population individuals were compared between datasets containing phenotypic records measured in the whole range of temperatures (WR) and datasets containing only phenotypic records measured when the weather station temperature was above 10°C (10C) or 15°C (15C), to evaluate the usefulness of these datasets that may better reflect the within-barn temperature. The use of homogeneous or heterogeneous residual variance was found to be trait-dependent, where homogeneous variance presented the best fit for MDP, BFT, OTW, TNB, NBA, WN and IBF, while the other traits (WW and IWE) had better fit with heterogeneous variance. The average prediction accuracy, dispersion and bias values considering all traits for WR were 0.36 ± 0.05, -0.07 ± 0.13 and 0.76 ± 0.10, respectively; for 10C were 0.39 ± 0.02, -0.05 ± 0.07 and 0.81 ± 0.05, respectively; and for 15C were 0.32 ± 0.05, -0.05 ± 0.11 and 0.84 ± 0.10, respectively. Based on the studied traits, using phenotypic records collected when the outside temperature (from public weather stations) was above 10°C provided better predictions for most of the traits. Forty-three and 62 candidate genomic regions were associated with the intercept (overall performance level) and slope term (specific biological mechanisms related to environmental sensitivity), respectively. Our results contribute to improve genomic predictions using existing datasets and better understand the genetic background of heat tolerance in pigs. Furthermore, the genomic regions and candidate genes identified will contribute to future genomic studies and breeding applications.

Recognition of human mood, alertness and comfort under the influence of indoor lighting using physiological features

Light is an essential and vital component of indoor environments, and it has both visual and non-visual effects on the human body. Illuminance and correlated colour temperature (CCT) have been reported to be closely related to human mood, alertness, and comfort. Recent research has focused on subjective feelings, test results, and transient physiological indicators; however, few studies have used continuous physiological data to explore the physiological features along with light parameters associated with subjective feelings. In this paper, we conducted light experiments at six illuminance levels of functional lighting (300, 400, 500, 600, 700, and 800 lx) and four typical CCT levels (3000, 4000, 5000, and 6000 K) to analyse the light effects on users’ mood, alertness, and visual comfort. The results showed that both illuminance and CCT significantly affected subjective perception. Notably, subjective feelings under the influence of the light environment were evaluated using the CatBoost method through 41 columns of physiological features extracted from electrocardiogram (ECG), electrodermal activity (EDA), skin temperature (SKT) and blood oxygen saturation (SpO2) data. The average accuracy values of the recognition framework for binary classification proposed in this study were 97.4% for the valence of mood, 95.7% for the arousal of mood, 97.1% for alertness, and 97.3% for visual comfort (all standard deviations were less than 2.5%). The research results can help intelligent buildings sense changes in human feelings and adjust control systems to construct human-centred indoor environments for work and study, as well as provide a reference of physiological features for personalised user experience research.

A Fully Privacy-Preserving Solution for Anomaly Detection in IoT using Federated Learning and Homomorphic Encryption

AbstractAnomaly detection for the Internet of Things (IoT) is a very important topic in the context of cyber-security. Indeed, as the pervasiveness of this technology is increasing, so is the number of threats and attacks targeting smart objects and their interactions. Behavioral fingerprinting has gained attention from researchers in this domain as it represents a novel strategy to model object interactions and assess their correctness and honesty. Still, there exist challenges in terms of the performance of such AI-based solutions. The main reasons can be alleged to scalability, privacy, and limitations on adopted Machine Learning algorithms. Indeed, in classical distributed fingerprinting approaches, an object models the behavior of a target contact by exploiting only the information coming from the direct interaction with it, which represents a very limited view of the target because it does not consider services and messages exchanged with other neighbors. On the other hand, building a global model of a target node behavior leveraging the information coming from the interactions with its neighbors, may lead to critical privacy concerns. To face this issue, the strategy proposed in this paper exploits Federated Learning to compute a global behavioral fingerprinting model for a target object, by analyzing its interactions with different peers in the network. Our solution allows the training of such models in a distributed way by relying also on a secure delegation strategy to involve less capable nodes in IoT. Moreover, through homomorphic encryption and Blockchain technology, our approach guarantees the privacy of both the target object and the different workers, as well as the robustness of the strategy in the presence of attacks. All these features lead to a secure fully privacy-preserving solution whose robustness, correctness, and performance are evaluated in this paper using a detailed security analysis and an extensive experimental campaign. Finally, the performance of our model is very satisfactory, as it consistently discriminates between normal and anomalous behaviors across all evaluated test sets, achieving an average accuracy value of 0.85.

QUALITY CONTROL CT SCAN (ANALISIS DAN EVALUASI KUALITAS CITRA)

Producing high-quality CT scans is crucial for accurately interpreting images and obtaining maximum diagnostic information. However, it usually comes at the cost of increased radiation exposure for the patient. Therefore, it's essential to evaluate CT scan image quality to ensure that it provides maximum information for diagnostic purposes while keeping the radiation dose received by the patient as low as possible. Several parameters can be used to evaluate the quality of CT scan images, such as accuracy, linearity, uniformity of the CT number, and high contrast resolution (spatial resolution). A phantom scan is carried out routinely once a week to test the quality of CT scan images. In this study, the image quality phantom was scanned using a Philips MX Evo 16 slice CT Scan over three months, which resulted in twelve CT scan image quality phantom image scan data. The CT scan data was processed and analyzed automatically using the IndoQCT application to determine the image quality. The value obtained was then compared with the test pass value stipulated in the Regulation of the Nuclear Energy Regulatory Agency (BAPETEN) Number 2 of 2018 concerning Compliance Tests for Diagnostic and Interventional Radiology X-ray. The average CT number accuracy value obtained was 0.6 HU, with a maximum value of 0.9 HU and a minimum value of 0.2 HU. The average CT number uniformity value obtained was 0.55 HU with a maximum value of 0.8 HU. The average CT number linearity value obtained was 0.9957 HU with a minimum of 0.9956. Meanwhile, the average MTF value obtained was 0.65, with a maximum value of 0.72. This study concludes that the values obtained still meet the CT Scan image quality test set by BAPETEN. Therefore, the CT Scan can still be considered reliable and used for service. Keywords: CT Scan, image quality, CT number, spatial resolution, IndoQCT.

PERFORMANCE OF FACTOR RETENTION METHODS IN SKEWED DISTRIBUTIONS

This research aims to evaluate the performance of dimensionality determination methods under various simulation conditions. Therefore, dimensionality determination methods were compared, including optimal parallel analysis, MAP, HULL, EGA (TMFG) estimation, EGA (glasso) estimation, and comparison data forest method. The type of distribution, sample size, number of items per factor, number of categories, and measurement model were specified as simulation conditions in the study. For each condition, 100 replications were conducted. A fully crossed simulation design was employed in the study. The results of this study, which examined the performance of factor determination methods under skewed distributions, indicated that the HULL method had the highest average considering the average accuracy values of all conditions. Meanwhile, the HULL method had the lowest RB average. However, no method demonstrated adequate performance under all conditions. This study examined one-factor and two-factor structures with interfactor correlations of 0.00 and 0.30. Considering structures with more than two factors in education and psychology, future research could focus on working with data exhibiting skewed distributions involving more factors and items to compare the performance of methods.

Research on Named Entity Recognition Method of Chinese Classics Under the Supervision of Domain Knowledge

The current dominant named entity recognition methods of Chinese classics are classified as data-driven methods, which are limited by the data quality. The domain knowledge is introduced in this paper to supervise the process of the named entity recognition, so as to solve the poor performance problem because of the low-quality data. The experiments on the Historical Records corpus show that compared with the domain knowledge unsupervised case, the average accuracy, recall rate, and F1 value have respectively improved by 2.76%, 2.70%, and 2.75% under the supervision of domain knowledge. Domain knowledge plays an important role in improving the performance of the named entity recognition methods of Chinese classics.

Neuroevolved bi-directional LSTM applied to zero emission zones management in urban transport

Plugin hybrid electric buses (PHEB) are nowadays an essential technology for the decarbonization of cities, proposing a more flexible solution than fully electric buses. There are a few studies for the design of operational strategies for PHEBs (i.e., to choose the propulsion method to use along the route) to minimize tailpipe emissions and/or maximize the electric range of the bus. However, all of them offer static strategies that cannot dynamically adapt to unforeseen traffic events, which can significantly impact on the performance of the bus. In this work, we propose modeling the problem of finding optimal PHEB operational strategies as a classification problem, allowing the generation of pseudo-optimal strategies in very short computational times. The method designed, called ML-EPBO, takes decisions based on local information of the bus, as well as from the route, so it can adapt to any events causing unexpected traffic conditions. Neuroevolution is used in this work to fine-tune ML-EPBO for optimal performance. Results show how ML-EPBO instantly offer solutions highly competitive with the best existing techniques in the literature, which do not offer the possibility of dynamically adapting to unexpected traffic events because they make use of global knowledge of the problem and have high computational complexity. Indeed, the average accuracy value obtained by the model is over 99.2%, and the strategies proposed are predicted over 70% of the cases with 100% of accuracy, even outperforming the state of the art results by up to 4.7% in the level of emissions or 1.82% in the electric range.

Prediction of diabetes disease using an ensemble of machine learning multi-classifier models

Background and objectiveDiabetes is a life-threatening chronic disease with a growing global prevalence, necessitating early diagnosis and treatment to prevent severe complications. Machine learning has emerged as a promising approach for diabetes diagnosis, but challenges such as limited labeled data, frequent missing values, and dataset imbalance hinder the development of accurate prediction models. Therefore, a novel framework is required to address these challenges and improve performance.MethodsIn this study, we propose an innovative pipeline-based multi-classification framework to predict diabetes in three classes: diabetic, non-diabetic, and prediabetes, using the imbalanced Iraqi Patient Dataset of Diabetes. Our framework incorporates various pre-processing techniques, including duplicate sample removal, attribute conversion, missing value imputation, data normalization and standardization, feature selection, and k-fold cross-validation. Furthermore, we implement multiple machine learning models, such as k-NN, SVM, DT, RF, AdaBoost, and GNB, and introduce a weighted ensemble approach based on the Area Under the Receiver Operating Characteristic Curve (AUC) to address dataset imbalance. Performance optimization is achieved through grid search and Bayesian optimization for hyper-parameter tuning.ResultsOur proposed model outperforms other machine learning models, including k-NN, SVM, DT, RF, AdaBoost, and GNB, in predicting diabetes. The model achieves high average accuracy, precision, recall, F1-score, and AUC values of 0.9887, 0.9861, 0.9792, 0.9851, and 0.999, respectively.ConclusionOur pipeline-based multi-classification framework demonstrates promising results in accurately predicting diabetes using an imbalanced dataset of Iraqi diabetic patients. The proposed framework addresses the challenges associated with limited labeled data, missing values, and dataset imbalance, leading to improved prediction performance. This study highlights the potential of machine learning techniques in diabetes diagnosis and management, and the proposed framework can serve as a valuable tool for accurate prediction and improved patient care. Further research can build upon our work to refine and optimize the framework and explore its applicability in diverse datasets and populations.

High-Level K-Nearest Neighbors (HLKNN): A Supervised Machine Learning Model for Classification Analysis

The k-nearest neighbors (KNN) algorithm has been widely used for classification analysis in machine learning. However, it suffers from noise samples that reduce its classification ability and therefore prediction accuracy. This article introduces the high-level k-nearest neighbors (HLKNN) method, a new technique for enhancing the k-nearest neighbors algorithm, which can effectively address the noise problem and contribute to improving the classification performance of KNN. Instead of only considering k neighbors of a given query instance, it also takes into account the neighbors of these neighbors. Experiments were conducted on 32 well-known popular datasets. The results showed that the proposed HLKNN method outperformed the standard KNN method with average accuracy values of 81.01% and 79.76%, respectively. In addition, the experiments demonstrated the superiority of HLKNN over previous KNN variants in terms of the accuracy metric in various datasets.

Software Subclassification Based on BERTopic-BERT-BiLSTM Model

With the continuous influx of application software onto the application software market, achieving accurate software recommendations for users in the huge software application market is urgent. To address this issue, each application software market currently provides its own classification tags. However, several problems still exist, such as the lack of objectivity, hierarchy, and standardization in these classifications, which in turn affects the accuracy of precise software recommendations. Accordingly, a customized BERTopic model is proposed to cluster the software description texts of the application software and the automatic tagging and updating of the application software tags are realized according to the clusters obtained by topic clustering and the extracted subject words. At the same time, a data enhancement method based on the c-TF-IDF algorithm is proposed to solve the problem of imbalance of datasets, and then the classification model based on the BERT-BiLSTM model is trained on the labeled datasets to classify the software in the dimension of the application function, so as to realize the accurate software recommendation for users. Based on the experimental verification of two datasets, 21 categories in the SourceForge dataset and 19 categories in the Chinese App Store dataset are subclassed by the clustering results of the customized BERTopic model, and the tags of 138 subclasses and 262 subclasses are formed, respectively. In addition, a complete tagged software description text dataset is constructed and the software tags are updated automatically. In the first stage of the classification experiment, the weighted average accuracy, recall rate, and F1 value can reach 0.92, 0.91, and 0.92, respectively. In the second stage, the weighted average accuracy, recall rate, and F1 value can all reach 0.96. After data enhancement, the weighted average F1 value of the classification model can be increased by up to two percentage points.

Intelligent recognition of greenhouse cucumber canopy vine top with deep learning model

In order to solve the problem of low recognition accuracy of cucumber canopy vine tops image caused by the high density and intermingling of the cucumber canopy vine, an image recognition model of cucumber canopy tops based on the deep learning model of improved YOLOv5 is proposed in this paper. A combination of online and offline augmentation is used to amplify the original data to ensure a balanced number of images in different types of datasets during training and testing. Considering the small leaves characteristics of the cucumber canopy vine tops, the CA mechanism module (YOLOv5-CA) is introduced into the backbone network of the YOLOv5 model to improve the recognition accuracy of the small leaf target. On this basis, the regression Loss GIOU of the target position was changed to EIOU and the cross-entropy function of confidence loss was changed to Focal Loss function (YOLOv5-CA-LS) to reduce the influence of the imbalance of difficult and easy samples on the detection results. The models before and after optimization were compared. The results show that compared with the YOLOv5 model, the accuracy, recall rate, mAP@0.5 and mAP@0.5:0.95 of the YOLOv5-CA-LS model is improved by 4.4%, 5.0%, 1.3% and 1.8% respectively. And average recognition accuracy value of the instance of the cucumber canopy tops adopted the YOLOv5-CA-LS model is 97.1% under different angle and cloudy conditions, which can meet the recognition requirement of high density and intermingling of the cucumber canopy vine tops.