Average Recall Values Research Articles

A novel software defect prediction approach using modified objective cluster analysis

AbstractA novel approach for software defect prediction of unlabeled datasets is proposed using modified objective cluster analysis (OCA). In this approach, the first step is to construct the distance matrix of instances in the datasets by utilizing the automatically determined clusters through the modified OCA. Then the dipoles within different instances are categorized into two different groups. Finally, the clusters of instances are produced, and software defects can be predicted by imposing a modified consistency criterion. Case study and comparative experiments were conducted based on 12 public datasets selected from the databases of Promise and ReLink using multiple different unsupervised algorithms and cross‐project approaches. There are two experimental settings: experiments with datasets that contain all metrics and experiments with datasets that contain only module size metrics. The results were evaluated by the index of precision, recall, F‐measure, and receiver operating characteristic curve (AUC). Furthermore, a complexity analysis of the tested algorithms was conducted as well. In experiments with datasets with all metrics, the proposed OCA gets the best results in four indexes, and the average values of precision, recall, F‐measure, and AUC were improved by a minimum of 1.52%, 2.78%, 19.84%, and 0.93%, respectively. In experiments with datasets with only module size metrics, the proposed OCA performed the best results in four indexes also, and the average values of precision, F‐measure, and AUC were improved by a minimum of 8.8%, 2.59%, and 8.36%, respectively. The proposed algorithm is of low complexity and provides a new way to efficiently predict software defects with unlabeled datasets.

Detecting COVID-19 patients based on fuzzy inference engine and Deep Neural Network

COVID-19, as an infectious disease, has shocked the world and still threatens the lives of billions of people. Recently, the detection of coronavirus (COVID-19) is a critical task for the medical practitioner. Unfortunately, COVID-19 spreads so quickly between people and approaches millions of people worldwide in few months. It is very much essential to quickly and accurately identify the infected people so that prevention of spread can be taken. Although several medical tests have been used to detect certain injuries, the hopefully detection efficiency has not been accomplished yet. In this paper, a new Hybrid Diagnose Strategy (HDS) has been introduced. HDS relies on a novel technique for ranking selected features by projecting them into a proposed Patient Space (PS). A Feature Connectivity Graph (FCG) is constructed which indicates both the weight of each feature as well as the binding degree to other features. The rank of a feature is determined based on two factors; the first is the feature weight, while the second is its binding degree to its neighbors in PS. Then, the ranked features are used to derive the classification model that can classify new persons to decide whether they are infected or not. The classification model is a hybrid model that consists of two classifiers; fuzzy inference engine and Deep Neural Network (DNN). The proposed HDS has been compared against recent techniques. Experimental results have shown that the proposed HDS outperforms the other competitors in terms of the average value of accuracy, precision, recall, and F-measure in which it provides about of 97.658%, 96.756%, 96.55%, and 96.615% respectively. Additionally, HDS provides the lowest error value of 2.342%. Further, the results were validated statistically using Wilcoxon Signed Rank Test and Friedman Test.

Content-based recommender system architecture for similar e-commerce products

Recommendation systems are quite famous and are increasingly being used on e-commerce platforms for a variety of purposes. The recommendation system technique used also varies greatly depending on the scope and Item of recommendation. Content-based filtering, for example, is used to recommend related product items based on user preferences. However, how the recommendation system architecture should be built starts by creating a data model for bringing up related product items. This paper offers a system architecture by considering the initial problem usually faced by recommendation systems, namely the cold start problem. The problem of lack of user preferences data is trying to be overcome by utilizing product item documents. Product item documents are processed using the TF-IDF algorithm and Vector Space Model to generate a data model. Then a query can be applied to find similarities to items that the user has seen. In the end, the recommendation system architecture that was built produced excellent Precision using Recall and Precision testing. Tests are carried out for data using the weighting of product names and product labels. The result obtained 0.84 for the average value of Recall and 0.78 for the average value of Precision.

Accelerating Automated Stomata Analysis Through Simplified Sample Collection and Imaging Techniques.

Digital image processing is commonly used in plant health and growth analysis, aiming to improve research efficiency and repeatability. One focus is analysing the morphology of stomata, with the aim to better understand the regulation of gas exchange, its link to photosynthesis and water use and how they are influenced by climatic conditions. Despite the key role played by these cells, their microscopic analysis is largely manual, requiring intricate sample collection, laborious microscope application and the manual operation of a graphical user interface to identify and measure stomata. This research proposes a simple, end-to-end solution which enables automatic analysis of stomata by introducing key changes to imaging techniques, stomata detection as well as stomatal pore area calculation. An optimal procedure was developed for sample collection and imaging by investigating the suitability of using an automatic microscope slide scanner to image nail polish imprints. The use of the slide scanner allows the rapid collection of high-quality images from entire samples with minimal manual effort. A convolutional neural network was used to automatically detect stomata in the input image, achieving average precision, recall and F-score values of 0.79, 0.85, and 0.82 across four plant species. A novel binary segmentation and stomatal cross section analysis method is developed to estimate the pore boundary and calculate the associated area. The pore estimation algorithm correctly identifies stomata pores 73.72% of the time. Ultimately, this research presents a fast and simplified method of stomatal assay generation requiring minimal human intervention, enhancing the speed of acquiring plant health information.

Automated medical diagnosis of COVID-19 through EfficientNet convolutional neural network

COVID-19 infection was reported in December 2019 at Wuhan, China. This virus critically affects several countries such as the USA, Brazil, India and Italy. Numerous research units are working at their higher level of effort to develop novel methods to prevent and control this pandemic scenario. The main objective of this paper is to propose a medical decision support system using the implementation of a convolutional neural network (CNN). This CNN has been developed using EfficientNet architecture. To the best of the authors’ knowledge, there is no similar study that proposes an automated method for COVID-19 diagnosis using EfficientNet. Therefore, the main contribution is to present the results of a CNN developed using EfficientNet and 10-fold stratified cross-validation. This paper presents two main experiments. First, the binary classification results using images from COVID-19 patients and normal patients are shown. Second, the multi-class results using images from COVID-19, pneumonia and normal patients are discussed. The results show average accuracy values for binary and multi-class of 99.62% and 96.70%, respectively. On the one hand, the proposed CNN model using EfficientNet presents an average recall value of 99.63% and 96.69% concerning binary and multi-class, respectively. On the other hand, 99.64% is the average precision value reported by binary classification, and 97.54% is presented in multi-class. Finally, the average F1-score for multi-class is 97.11%, and 99.62% is presented for binary classification. In conclusion, the proposed architecture can provide an automated medical diagnostics system to support healthcare specialists for enhanced decision making during this pandemic scenario.

BUILDING OUTLINE DELINEATION: FROM VERY HIGH RESOLUTION REMOTE SENSING IMAGERY TO POLYGONS WITH AN IMPROVED END-TO-END LEARNING FRAMEWORK

Abstract. Deep learning methods based on Fully convolution networks (FCNs) have shown an impressive progress in building outline delineation from very high resolution (VHR) remote sensing (RS) imagery. Common issues still exist in extracting precise building shapes and outlines, often resulting in irregular edges and over smoothed corners. In this paper, we use PolyMapper, a recently introduced deep-learning framework that is able to predict object outlines in a vector representation directly. We have introduced two main modifications to this baseline method. First, we introduce EffcientNet as backbone feature encoder to our network, which uses compound coefficient to scale up all dimensions of depth/width/resolution uniformly, to improve the processing speed with fewer parameters. Second, we integrate a boundary refinement block (BRB) to strengthen the boundary feature learning and to further improve the accuracy of corner prediction. The results demonstrate that the end-to-end learnable model is capable of delineating polygons of building outlines that closely approximate the structure of reference labels. Experiments on the crowdAI building instance segmentation datasets show that our model outperforms PolyMapper in all COCO metrics, for instance showing a 0.13 higher mean Average Precision (AP) value and a 0.60 higher mean Average Recall value. Also qualitative results show that our method segments building instances of various shapes more accurately.

Scene text detection using enhanced Extremal region and convolutional neural network

Text in scene images usually contains significant information. Text detection and recognition in the scene is important for a variety of advanced machine vision applications, such as image and video retrieval, automotive assistance, and multilingual translation. In particular, most text recognition systems require texts to be localized in images beforehand and this is a significant demand. The purpose of this study is to provide a method to detect texts in natural images. The proposed approach combines advantages of extremal region, ER, methods and classification of convolutional neural network, CNN. This significantly reduces the false positives and increases the accuracy of detection. The method of sliding windows is employed with different sizes in order to determine text candidates. Extraction of enhanced ERs is performed in three consecutive stages on three distinct color channels, R, G, and B. Then, the results are combined together by an add method. After grouping, the word candidates are classified to two classes of text and non-text sections by a CNN classifier. By applying non-maximum suppression (NMS) algorithm to the same words, words with the highest probability are selected. The average values of accuracy, recall, precision and F-measure of the proposed text detection model on the ICDAR2013 database are 0.893, 0.962, 0.948, and 0.955, respectively. The optimal cut point of the proposed method is 0.648, which has the highest average accuracy, 91.93%. The AUC of ROC and PR diagrams for the proposed model are 0.851 and 0.718, respectively. These results of AUC for ROC and PR curves showed an outstanding enhancement in comparison with the best detection rate of previous methods. Experimental results on the ICDAR2011, ICDAR2013 and ICDAR2015 databases also demonstrate that our algorithm outperforms the state-of-the-art scene text detection methods.

Multilabel Classification for News Article Using Long Short-Term Memory

Multilabel text classification is a task of categorizing text into one or more categories. Like other machine learning, multilabel classification performance is limited when there is small labeled data and leads to the difficulty of capturing semantic relationships. In this case, it requires a multi-label text classification technique that can group four labels from news articles. Deep Learning is a proposed method for solving problems in multi-label text classification techniques. By comparing the seven proposed Long Short-Term Memory (LSTM) models with large-scale datasets by dividing 4 LSTM models with 1 layer, 2 layer and 3-layer LSTM and Bidirectional LSTM to show that LSTM can achieve good performance in multi-label text classification. The results show that the evaluation of the performance of the 2-layer LSTM model in the training process obtained an accuracy of 96 with the highest testing accuracy of all models at 94.3. The performance results for model 3 with 1-layer LSTM obtained the average value of precision, recall, and f1-score equal to the 94 training process accuracy. This states that model 3 with 1-layer LSTM both training and testing process is better. The comparison among seven proposed LSTM models shows that model 3 with 1 layer LSTM is the best model.

Vehicle Detection and Counting Using Adaptive Background Model Based on Approximate Median Filter and Triangulation Threshold Techniques

Background subtraction method is widely used for vehicle detection. One of the issues in this method is to find a suitable and accurate background model that works in all conditions. Moreover, setting an appropriate threshold value to discriminate between the moving objects and stationary background plays a crucial role in increasing the detection performance. In this paper, an adaptive background model combined with an adaptive threshold method is proposed. It is demonstrated that the proposed method can efficiently differentiate between moving vehicles and background in urban roads under different weather conditions (i.e., normal, rainy, foggy, and snowy). Comparisons between the proposed method and other methods, such as the adaptive local threshold (ALT) and the three frame-differencing methods show the potential of our approach. The experimental results show that the proposed method increases the average recall value by 16.4% and the average precision value by 21.7% in comparison to the ALT method with a negligible increase in the processing time.

Classification of Normal and Cardiomegaly Conditions in Chest Radiographs Using Cardio-Mediastinal Features.

In this study, an attempt has been made to differentiate normal and cardiomegaly using cardio-mediastinal ratiometric features and machine learning approaches. A total of 60 chest radiographs including normal and cardiomegaly subjects are considered from a public dataset. The images are preprocessed using edge aware contrast enhancement technique to improve the edge contrast of lung boundaries. The mediastinal, cardiac and thoracic widths and their ratiometric indices are computed to characterize the morphological variations. The features are fed to three different classifiers for the differentiation of normal and cardiomegaly. Results show that the Linear discriminant analysis classifier is found to perform better with average values of recall 88.7%, precision 88.8%, and area under the curve 91.9%. Hence, the proposed computer aided diagnostic approach appears to be clinically significant to distinguish normal and cardiomegaly especially in remote and resource - poor settings.

DeepNerve: A New Convolutional Neural Network for the Localization and Segmentation of the Median Nerve in Ultrasound Image Sequences

Carpal tunnel syndrome commonly occurs in individuals working in occupations that involve use of vibrating manual tools or tasks with highly repetitive and forceful manual exertion. In recent years, carpal tunnel syndrome has been evaluated by ultrasound imaging that monitors median nerve movement. Conventional image analysis methods, such as the active contour model, are typically used to expedite automatic segmentation of the median nerve, but these usually suffer from an arduous manual intervention. We propose a new convolutional neural network framework for localization and segmentation of the median nerve, called DeepNerve, that is based on the U-Net model. DeepNerve integrates the characteristics of MaskTrack and convolutional long short-term memory to effectively locate and segment the median nerve. On the basis of experimental results, the proposed model achieved high performance and generated average Dice measurement, precision, recall and F-score values of 0.8975, 0.8912, 0.9119 and 0.9015, respectively. The segmentation results of DeepNerve were significantly improved in comparison with those of conventional active contour models. Additionally, the results of Student's t-test revealed significant differences in four deformation measurements of the median nerve, including area, perimeter, aspect ratio and circularity. We conclude that the proposed DeepNerve not only generates satisfactory results for localization and segmentation of the median nerve, but also creates more promising measurements for applications in clinical carpal tunnel syndrome diagnosis.

Representing cancer clinical trial criteria and attributes using ontologies: An NLP-assisted approach.

e14079 Background: Low patient recruitment is one of the main reasons clinical trials fail. Identifying eligible patients for clinical trials using electric health records (EHRs) can help reach accrual targets. Ontology reasoning implemented in Trial2Patient, a scalable system we developed for matching patient to clinical trials, forms the basis for generating patient cohorts in our system. For efficient cohort definition, an attribute ontology for eligibility criteria and entity categorization is a necessary first step. To meet this requirement, we constructed an ontology platform for lung cancer trials. Methods: We classified 128 non-small cell lung cancer and 38 small cell lung cancer trials into different therapy groups. Among the 166 trials we examined, 110 were immuno-oncology therapy-based, 48 were targeted therapy-based, and 8 were chemotherapy or device trials. We analyzed the eligibility criteria for each trial manually to identify entities from all trials as well as indication specific and further therapy group specific entities. To incorporate a semi-automated, natural language process (NLP)-assisted named entity recognition (NER) into the future cohort definition process, we trained NLP and deep learning models for NER and ontology encoding. Attributes generated from 50 processed NSCLC trials were evaluated with our manually curated attributes. The ontology generated from lung cancer was tested in 74 prostate cancer trials for generalizability. Results: The ontology for lung cancer trials, which is generalizable to prostate cancer and other cancer clinical trials, were constructed. Total 507 attributes were extracted and entities were categorized into 8 groups. Evaluation of attributes generated by NLP and deep learning models compared with manually extracted attributes showed high consistency and accuracy. The average precision, recall and F1 values of 15 most commonly appearing entities (disease, histology, targeted therapy, immunotherapy, radiotherapy, neoadjuvant therapy, age, gender, test, vitals, value, drug, gene, mutation, problem) are 0.873, 0.769, and 0.805, respectively. Conclusions: We contribute to a clinical trial ontology platform for lung cancer and prostate cancer trial recruitment. This ontology platform can be expanded to other solid tumors or hematologic malignancies for clinical trial analysis, and can also be applied to generate synthetic control arm cohorts. We believe NLP-assisted NER can be successfully incorporated for the future work of large scale of clinical trial cohort definition.

Source Code Assessment and Classification Based on Estimated Error Probability Using Attentive LSTM Language Model and Its Application in Programming Education

The rate of software development has increased dramatically. Conventional compilers cannot assess and detect all source code errors. Software may thus contain errors, negatively affecting end-users. It is also difficult to assess and detect source code logic errors using traditional compilers, resulting in software that contains errors. A method that utilizes artificial intelligence for assessing and detecting errors and classifying source code as correct (error-free) or incorrect is thus required. Here, we propose a sequential language model that uses an attention-mechanism-based long short-term memory (LSTM) neural network to assess and classify source code based on the estimated error probability. The attentive mechanism enhances the accuracy of the proposed language model for error assessment and classification. We trained the proposed model using correct source code and then evaluated its performance. The experimental results show that the proposed model has logic and syntax error detection accuracies of 92.2% and 94.8%, respectively, outperforming state-of-the-art models. We also applied the proposed model to the classification of source code with logic and syntax errors. The average precision, recall, and F-measure values for such classification are much better than those of benchmark models. To strengthen the proposed model, we combined the attention mechanism with LSTM to enhance the results of error assessment and detection as well as source code classification. Finally, our proposed model can be effective in programming education and software engineering by improving code writing, debugging, error-correction, and reasoning.

QUERY ANSWERING SYSTEM OF SHAHIH HADITH MUTTAFAQUN ‘ALAIH USING INDONESIAN THESAURUS BASED ON QUERY EXPANSION AND NAÏVE BAYES CLASSIFIER

<p class="Text"><em><span style="font-size: 9.0pt; line-height: 105%;">Abstract</span></em><span style="font-size: 9.0pt; line-height: 105%;">— Hadith are all the words, deeds and provisions of the Prophet Muhammad SAW that are used as the second of Islamic law after Al-Quran. The purpose of this study is to make an Information Retrieval system called the Query Answering System is expected to facilitate users in searching and finding the hadith documents as the user's needs. This study implements the Naïve Bayes Classifier method combined with Indonesian thesaurus as a query expansion to find the hadith documents that relevant to the input query. Based on the testing of 50 query data, the test results show that the use of query expansion gives better results than without using query expansion. Where based on testing of the top 1 data without using query expansion obtained an average recall value of 62%, an average precision value of 62%, an average accuracy value of 92.4% and an average value of the f-measure of 62%, while testing using query expansion obtained an average recall value of 66%, an average precision value of 66%, an average accuracy value of 93.2% and an average f -measure value of 66%. Based on the test results, the use of query expansion shows an improvement in the average recall value of 4%, an improvement in the average precision value of 4%, and an improvement in the average accuracy value of 0.8% and an improvement in the average f-measure value of 4% compared on without using query expansion.</span></p><p class="MsoNormal"> </p><p class="IndexTerms"><em>Index Terms</em>—hadith, information retrieval, query expansion, naïve bayes. </p>

A Novel Approach for Software Defect prediction Based on the Power Law Function

Power law describes a common behavior in which a few factors play decisive roles in one thing. Most software defects occur in very few instances. In this study, we proposed a novel approach that adopts power law function characteristics for software defect prediction. The first step in this approach is to establish the power law function of the majority of metrics in a software system. Following this, the power law function’s maximal curvature value is applied as the threshold value for determining higher metric values. Furthermore, the total number of higher metric values is counted in each instance. Finally, the statistical data are clustered into different categories as defect-free and defect-prone instances. Case studies and a comparison were conducted based on twelve public datasets of Promise, SoftLab, and ReLink by using five different algorithms. The results indicate that the precision, recall, and F-measure values obtained by the proposed approach are the most optimal among the tested five algorithms, the average values of recall and F-measure were improved by 14.3% and 6.0%, respectively. Furthermore, the complexity of the proposed approach based on the power law function is O ( 2 n ) , which is the lowest among the tested five algorithms. The proposed approach is thus demonstrated to be feasible and highly efficient at software defect prediction with unlabeled datasets.

Counting of Bacterial Colonies of the Low Quality Image Using Perona-Malik Diffusion Filters and Image Morphology Operators

Microbiology has an important role in solving human problems that are complex in many fields, such as environment, medical, and biotechnology. One of the important researches in microbiology fields is the counting of the number of bacterial colonies. This research is very important to estimate the number of bacterial cells in each millilitre or gram sample. However, the manual counting of bacterial colonies is a time-consuming task and is also very tedious. To help microbiologists, an automatic counting method of bacterial colonies based on bacterial colonies images is needed to be developed. However, the bacterial colonies images taking by the phones camera has low-quality. They are noise presence and lack of contrast, so the pre-processing method is necessary to be developed. This paper proposes the pre-processing method by using modified Perona-Malik Diffusion (PMD) and Contrast Limited Adaptive Histogram Equalization (CLAHE). The modified PMD filter is used to reduce noise, while the CLAHE method is used to enhance the contrast. Moreover, the morphology operators are used for the petri dish extraction and bacterial colonies counting process. The experimental results show that the proposed method has better performance than the original PMD based on the average values of precision, recall, and F-measure.

Benefits of ensemble models in road pavement cracking classification

AbstractThe maintenance of road pavements is an essential task to prevent major deterioration and to reduce accident rates. In this task, the detection and classification of different types of cracks on the roads is usually considered. However, in most cases, these tasks are not fully automated and they need to be supervised by an expert to make repair decisions. This work focuses on the automatic classification of the most common types of cracks: longitudinal cracks, transverse cracks, and alligator cracks. Our proposal combines, first, computer vision techniques for crack segmentation and second, an ensemble model (composed of different rule‐based algorithms) for the classification. This approach achieves an average precision and recall values greater than 94% for three analyzed data sets improving the results in comparison to other approaches.

Enhancing Failure Mode and Effects Analysis Using Auto Machine Learning: A Case Study of the Agricultural Machinery Industry

In this paper, multiclass classification is used to develop a novel approach to enhance failure mode and effects analysis and the generation of risk priority number. This is done by developing four machine learning models using auto machine learning. Failure mode and effects analysis is a technique that is used in industry to identify possible failures that may occur and the effects of these failures on the system. Meanwhile, risk priority number is a numeric value that is calculated by multiplying three associated parameters namely severity, occurrence and detectability. The value of risk priority number determines the next actions to be made. A dataset that includes a one-year registry of 1532 failures with their description, severity, occurrence, and detectability is used to develop four models to predict the values of severity, occurrence, and detectability. Meanwhile, the resulted models are evaluated using 10% of the dataset. Evaluation results show that the proposed models have high accuracy whereas the average value of precision, recall, and F1 score are in the range of 86.6–93.2%, 67.9–87.9%, 0.892–0.765% respectively. The proposed work helps in carrying out failure mode and effects analysis in a more efficient way as compared to the conventional techniques.

Latent semantic analysis and cosine similarity for hadith search engine

Search engine technology was used to find information as needed easily, quickly and efficiently, including in searching the information about the hadith which was a second guideline of life for muslim besides the Holy Qur'an. This study was aim to build a specialized search engine to find information about a complete and eleven hadith in Indonesian language. In this research, search engines worked by using latent semantic analysis (LSA) and cosine similarity based on the keywords entered. The LSA and cosine similarity methods were used in forming structured representations of text data as well as calculating the similarity of the keyword text entered with hadith text data, so the hadith information was issued in accordance with what was searched. Based on the results of the test conducted 50 times, it indicated that the LSA and cosine similarity had a success rate in finding high hadith information with an average recall value was 87.83%, although from all information obtained level of precision hadith was found semantically not many, it was indicated by the average precision value was 36.25%.

Analisis Klasifikasi Kanker Payudara Menggunakan Algoritma Naive Bayes

Breast cancer is one of the highest causes of death among women, this disease ranks second cause of death after lung cancer. According to the world health organization, 1 million women get a diagnosis of breast cancer every year and half of them die, in general this is due to early treatment and slow treatment resulting in new cancers being detected after entering the final stage. In the field of health and medicine, machine learning-based classification has been carried out to help doctors and health professionals in classifying the types of cancer, to determine which treatment measures should be performed. In this study breast cancer classification will be carried out using the Naive Bayes algorithm to group the types of cancer. The dataset used is from the Wisconsin breast cancer database. The results of this study are the ability of the Naive Bayes algorithm for the classification of breast cancer produces a good value, where the average percentage of correctly classified data reaches 96.9% and the average percentage of data is classified as incorrect only 3.1%. While the level of effectiveness of classification with naive bayes is high, where the average value of precision and recall is around 0.96. The highest precision and recall values are when the test data uses a percentage split of 40% with the respective values reaching 0.974 and 0.973.