Classifier System Research Articles

Toward Evaluating Critical Factors of Extubation Outcome with XCSR-Generated Rules.

Predicting the correct timing for extubation is pivotal for critically ill patients with mechanical ventilation support. Evidence suggests that extubation failure occurs in approximately 15-20% of patients, despite their passing of the extubation evaluation, necessitating reintubation. For critically ill patients, reintubation invariably increases mortality risk and medical costs. The numerous parameters that have been proposed for extubation decision-making, which constitute the key predictors of successful extubation, remains unclear. In this study, an extended classifier system capable of processing real-value inputs was proposed to select features of successful extubation. In total, 40 features linked to clinical information and variables acquired during spontaneous breathing trial (SBT) were used as the environmental inputs. According to the number of "don't care" rules in a population set, Probusage, the probability of the feature not being classified as above rules, can be calculated. A total of 228 subjects' results showed that Probusage was higher than 90% for minute ventilation at the 1st, 30th, 60th, and 90th minutes; respiratory rate at the 90th minute; and body weight, indicating that the variance in respiratory parameters during an SBT are critical predictors of successful extubation. The present XCSR model is useful to evaluate critical factors of extubation outcomes. Additionally, the current findings suggest that SBT duration should exceed 90 min, and that clinicians should consider the variance in respiratory variables during an SBT before making extubation decisions.

Application of Haar-like Features in Three AdaBoost Algorithms for Face Detection

In this paper we introduce a proposed cascade classifier system consists of five strong classifiers (stages), and empirically analysis three types of Adaboost algorithms (Real, Modest, and Gentle) which it were used for boosting the strong classifiers. Five prototypes of haar-like features (two-horizontal rectangles, two-vertical rectangles, three-horizontal rectangles, three-vertical rectangles, and four-rectangles) are used for each strong classifier (stage) of cascade classifier. Two-horizontal, two-vertical rectangles features and Modest-Adaboost Algorithm are applied for the 1st and 2nd strong classifier respectively, three-horizontal, vertical-rectangles features and Real-Adaboost Algorithm are applied for 3rd and 4th strong classifier respectively, and the last, four-rectangles and Real-Adaboost Algorithm are applied for the 5th strong classifier of cascade classifier. The implementation shows that the best use of Adaboost algorithm is: the Modest-Adaboost algorithm for both 1st and 2nd strong classifier, and the Real-Adaboost algorithm for 3rd,4th and 5th strong classifier.

Evolving multi-user fuzzy classifier system with advanced explainability and interpretability aspects

Evolving classifiers and especially evolving fuzzy classifiers have been established as a prominent technique for addressing the recent demands in building classifiers in an incremental online manner, based on target labels typically provided by a single user. We present a framework for an interactive evolving multi-user fuzzy classifier system with advanced explainability and interpretability aspects (EFCS-MU-AEI). Multiple users may provide their label feedback based on which own users’ classifiers are incrementally trained with evolving learning concepts. Its classification outputs are amalgamated by a specific ensembling scheme, respecting (i.) uncertainty in the class labels due to labeling ambiguities among the users and (ii.) different experience levels of the users as voting weights. A major focus thereby is concentrated on the explainability of classification outputs for the purpose to increase the quality (consistency and certainty) of the user (labeling) feedbacks. It is established to show reasons why certain decisions have been made and with which certainty levels and rule coverage degrees. The reasons are deduced from the most active rules, which are reduced in their length by a statistically-motivated instance-based feature importance level concept. Another major focus lies on the interpretability of extracted rules in order to represent understandable knowledge contained in the classification problem and especially to realize the labeling behaviors of different users for different parts of the feature space (= different sample groups). A specific incremental feature weighting technique, respecting label uncertainties from multiple users and sample forgetting weights (for handling drifts), as well as a fuzzy set merging process are proposed to aim for a high compactness and transparency of the rules.Our approach was evaluated based on a visual inspection scenario. It could be shown that the explanations of the classifier decisions in fact significantly improved the labeling behavior of three single users in terms of showing higher accumulated accuracy trends. Feature weights integration into the classifier updates could achieve transparent rules with final essential four features to describe the classification problem. Based on this description, it turned out in which ways, i.e. for which sample groups, the users with lower experience levels should be taught to improve their understanding about the process.

Increasing comprehensiveness and reducing workload in a systematic review of complex interventions using automated machine learning.

As part of our ongoing systematic review of complex interventions for the primary prevention of cardiovascular diseases, we have developed and evaluated automated machine-learning classifiers for title and abstract screening. The aim was to develop a high-performing algorithm comparable to human screening. We followed a three-phase process to develop and test an automated machine learning-based classifier for screening potential studies on interventions for primary prevention of cardiovascular disease. We labelled a total of 16,611 articles during the first phase of the project. In the second phase, we used the labelled articles to develop a machine learning-based classifier. After that, we examined the performance of the classifiers in correctly labelling the papers. We evaluated the performance of the five deep-learning models [i.e. parallel convolutional neural network ( CNN ), stacked CNN , parallel-stacked CNN , recurrent neural network ( RNN ) and CNN-RNN]. The models were evaluated using recall, precision and work saved over sampling at no less than 95% recall. We labelled a total of 16,611 articles, of which 676 (4.0%) were tagged as 'relevant' and 15,935 (96%) were tagged as 'irrelevant'. The recall ranged from 51.9% to 96.6%. The precision ranged from 64.6% to 99.1%. The work saved over sampling ranged from 8.9% to as high as 92.1%. The best-performing model was parallel CNN , yielding a 96.4% recall, as well as 99.1% precision, and a potential workload reduction of 89.9%. We used words from the title and the abstract only. More work needs to be done to look into possible changes in performance, such as adding features such as full document text. The approach might also not be able to be used for other complex systematic reviews on different topics. Our study shows that machine learning has the potential to significantly aid the labour-intensive screening of abstracts in systematic reviews of complex interventions. Future research should concentrate on enhancing the classifier system and determining how it can be integrated into the systematic review workflow. This project was funded by the National Institute for Health and Care Research (NIHR) Health Technology Assessment programme and will be published in Health Technology Assessment. See the NIHR Journals Library website for further project information.

Defining numeral classifiers and identifying classifier languages of the world

AbstractThis paper presents a precise definition of numeral classifiers, steps to identify a numeral classifier language, and a database of 3,338 languages, of which 723 languages have been identified as having a numeral classifier system. The database, named World Atlas of Classifier Languages (WACL), has been systematically constructed over the last 10 years via a manual survey of relevant literature and also an automatic scan of digitized grammars followed by manual checking. The open-access release of WACL is thus a significant contribution to linguistic research in providing (i) a precise definition and examples of how to identify numeral classifiers in language data and (ii) the largest dataset of numeral classifier languages in the world. As such it offers researchers a rich and stable data source for conducting typological, quantitative, and phylogenetic analyses on numeral classifiers. The database will also be expanded with additional features relating to numeral classifiers in the future in order to allow more fine-grained analyses.

Hand gestures recognition classification

The use of hand gestures for human-machine interaction offers an enticing alternative to bulky interface devices. The current study discusses the classification of gestures in real time and aims to create an algorithm capable of classifying gestural control commands accurately. For the classification of a gesture vocabulary of eight dynamic hand gestures, two separate classifiers were created. The established classifiers were: K-means + rule-based classifier and classifier of to test the accuracy of classification recognition in which a test set of 180 trajectories was categorized, an experiment was conducted. The accuracies obtained for the K-means and Learning classifier systems( LCS ) classifiers, respectively, are 90 and 94 percent.

Bayesian Classification of Nonmeteorological Targets in Polarimetric Doppler Radar Measurements

Abstract The latest established generation of weather radars provides polarimetric measurements of a wide variety of meteorological and nonmeteorological targets. While the classification of different precipitation types based on polarimetric data has been studied extensively, nonmeteorological targets have garnered relatively less attention beyond an effort to detect them for removal from meteorological products. In this paper we present a supervised learning classification system developed in the Finnish Meteorological Institute (FMI) that uses Bayesian inference with empirical probability density distributions to assign individual range gate samples into 7 meteorological and 12 nonmeteorological classes, belonging to five top-level categories of hydrometeors, terrain, zoogenic, anthropogenic, and immaterial. We demonstrate how the accuracy of the class probability estimates provided by a basic naive Bayes classifier can be further improved by introducing synthetic channels created through limited neighborhood filtering, by properly managing partial moment nonresponse, and by considering spatial correlation of class membership of adjacent range gates. The choice of Bayesian classification provides well-substantiated quality estimates for all meteorological products, a feature that is being increasingly requested by users of weather radar products. The availability of comprehensive, fine-grained classification of nonmeteorological targets also enables a large array of emerging applications, utilizing nonprecipitation echo types and demonstrating the need to move from a single, universal quality metric of radar observations to one that depends on the application, the measured target type, and the specificity of the customers’ requirements. Significance Statement In addition to meteorological echoes, weather radars observe a wide variety of nonmeteorological phenomena including birds, insects, and human-made objects like ships and aircraft. Conventionally, these data have been rejected as undesirable disturbance, but lately their value for applications like aeroecological monitoring of bird and insect migration has been understood. The utilization of these data, however, has been hampered by a lack of comprehensive classification of nonmeteorological echoes. In this paper we present a comprehensive, fine-grained, probabilistic classifier for all common types of nonmeteorological echoes which enables the implementation of a wide range of novel weather radar applications.

Design of intelligent garbage classifier system based on machine vision

With the rapid development of society, the amount of garbage produced in life is increasing day by day. How to deal with the garbage and how to classify the garbage scientifically and effectively are hot topics in the current society. To solve the above problems, this paper proposes an intelligent garbage classifier based on machine vision. The design uses STM32F103C8T6 single chip microcomputer as the core controller of the intelligent garbage collector. The hardware part of the system is composed of computer vision identification module, power module, photoelectric switch module, ultrasonic module, stepper motor driver and other modules. At the same time, different kinds of garbage are identified through the visual module. The single-chip microcomputer completes the software test of garbage classification through the stepper motor. Finally, the performance of the intelligent garbage collector is verified through experiments. The results show that the intelligent garbage classifier can realize the key functions of intelligent identification and automatic sorting and achieve high accuracy in the process of identifying different garbage. The design has great practical value.

OLP++: An online local classifier for high dimensional data

Ensemble diversity is an important characteristic of Multiple Classifier Systems (MCS), which aim at improving the overall performance of a classification system by combining the response of several models. While diversity may be introduced through various manipulations at the data level and the model level, some MCSs incorporate local information in order to increase it and/or take advantage of it, based on the idea that the different classifiers in the ensemble may have expertise in distinct areas of the feature space.11Code available at: https://github.com/marianaasouza/olp_plusplus.Following a similar reasoning, we introduced in a previous work an ensemble method which produces in test time a few experts in the local region where each given query sample is located. These local experts, which are generated with slightly differing views of the target area, are then used to label the corresponding unknown instance. While the framework was shown to perform well especially over imbalanced problems, the locality definition in the method is based on the nearest neighbors rule and Euclidian distance, as is the case of various local-based ensembles, which may suffer from the effects of the curse of dimensionality over high dimensional problems. Thus, in this work, we propose a local ensemble method in which we leverage the data partitions given by decision trees for locality definition. More specifically, the partitions defined at different levels of the decision path that a given query instance traverses in the tree(s) are used as the regions over which the local experts are produced. By using different node levels from the path, each classifier in the local pool has a moderately distinct view of the target region without resorting to a dissimilarity metric, which might be susceptible to high dimensional spaces. Experimental results over 39 high dimensional problems showed that the proposed approach was significantly superior to our previous, distance-based framework in balanced accuracy rate. Compared to other six local-based ensemble methods, including dynamic selection and weighting schemes, the proposed method achieved competitive results, outperforming the random forest baseline and two state-of-the-art dynamic ensemble selection techniques.

Exploring diversity in data complexity and classifier decision spaces for pool generation

This paper introduces a novel method for classifier pool generation in which a two-level strategy explores diversity in both data complexity and classifier decision spaces. The rationale is to induce pool members using data subsets representing subproblems with different difficulties while promoting diversity in classifiers’ decisions. Two possible variants of the proposed method with a focus on maximum dispersion and maximum accuracy are presented. These differ in the property used to define the best pool of classifiers provided by an optimization process. A robust experimental protocol encompassing 28 classification datasets shows that the proposed pool generation provided the best accuracy on 327 over 336 experiments (97.3%) when compared to well-known pool generation methods to provide multiple classifier systems with and without dynamic selection.

Correction: A Learning Classifier System for Automated Test Case Prioritization and Selection

Correction: A Learning Classifier System for Automated Test Case Prioritization and Selection

Water Quality Prediction and Classification based on Linear Discriminant Analysis and Light Gradient Boosting Machine Classifier Approach

Estimating water quality has existed as one of the vital factors embarked on the planet in the present eons. This paper illustrates a water quality estimate based on the Linear Discriminant Analysis (LDA) technique. Weighted arithmetic index technique is used in the computation of the Water Quality Index (WQI). At that moment, the LDA is linked to the dataset, and the ultimate principal WQI dynamics have been determined. Subsequently after predicting the WQI, Light Gradient Boosted Machine (LGBM) classification is performed in the LDA. Lastly, the LGBM classifier is activated to label the water quality. This proposed LGBM with LDA technique is demonstrated and evaluated on a Gulshan Lake-related dataset. The results show 96% forecast accuracy for the LDA and 100% categorization accuracy for the Light Gradient Boosted Machine classifier system that indicate consistent interpretation linked over the futuristic prototypes. This innovative model LDA-LGBM is aimed at enhancing the prediction of water quality and its classification through AI - ML approach.

AI BASED REALTIME FACE MASK DETECTION

The world has not yet fully Recover from this pandemic and the vaccine that can effectively treat Covid-19 is yet to be discovered. Wearing a face mask will help prevent the spread of infection and prevent the individual from contracting any airborne infectious germs. Done by evaluation of the classification results by analysing real-time streaming from the Camera. Using Face Mask Detection System, one can monitor if the people are wearing masks or not. Here HAARCASACADE algorithm is used for image detection. Collating with other existing algorithms, this classifier produces a high recognition rate even with varying expressions, efficient feature selection and low assortment of false positive features. HAAR feature-based cascade classifier system utilizes only 200 features out of 6000 features to yield a recognition rate of 98-100%. According to this motivation we demand mask detection as a unique and public health service system during the global pandemic COVID-19 epidemic. The model is trained by face mask image and non-face mask image. KEYWORDS Face mask detection Face edges (FMD) CNN- Convolutional Neural Network

Development and Application of an Intelligent Diagnosis System for Retinal Vein Occlusion Based on Deep Learning.

This study is aimed at developing an intelligent algorithm based on deep learning and discussing its application for the classification and diagnosis of retinal vein occlusions (RVO) using fundus images. A total of 501 fundus images of healthy eyes and patients with RVO were used for model training and testing to investigate an intelligent diagnosis system. The images were first classified into four categories by fundus disease specialists: (i) healthy fundus (group 0), (ii) branch RVO (BRVO) (group 1), (iii) central RVO (CRVO) (group 2), and (iv) macular branch RVO (MBRVO) (group 3), before being diagnosed using the ResNet18 network model. Intelligent diagnoses were compared with clinical diagnoses. The specificity of the intelligent diagnosis system under each attention mechanism was 100% in group 0 and also revealed a high sensitivity of over 95%, F1 score of over 97%, and an accuracy of over 97% in this group. For the other three groups, the specificities of diagnosis ranged from 0.45 to 0.91 with different attention mechanisms, in which the ResNet18+coordinate attention (CA) model had the highest specificities of 0.91, 0.88, and 0.83 for groups 1, 2, and 3, respectively. It also provided a high accuracy of over 94% with a coordinate attention mechanism in all four groups. The intelligent diagnosis and classifier system developed herein based on deep learning can determine the presence of RVO and classify disease according to the site of occlusion. This proposed system is expected to provide a new tool for RVO diagnosis and screening and will help solve the current challenges due to the shortage of medical resources.

Hierarchical Traffic Sign Recognition System

A hierarchical system of classifiers for recognizing traffic signs from their images is proposed in the article. Only signs that fit into a square are considered. The traffic signs set is analyzed by their shape, color and basic features. A hierarchy of classes for traffic signs on the details of their images is proposed. The traffic sign image recognition algorithm uses this hierarchy. The algorithm only works with localized signs. The localization algorithm is not considered. Image preprocessing is performed at each level of the hierarchy for traffic sign features. Different classifiers are used at different levels of the hierarchy. Preprocessing at the first level uses the segmentation method. Signs are classified by their form in the first level. The edge segment is highlighted in the image of the traffic sign. The Euclidean distance in color space is used to calculate the edge segment. The edge segment outline defines the shape of the sign. Classification of the sign by its shape is carried out based on comparison with the standards. The classifier defines the pattern closest to the sign contour. The efficiency of the classification in the first level is 99.8 %. The second level classifies signs by the color of the edge segment. The effectiveness of this classification is close to 100 %. Internal sign images are obtained after the edge segment is removed from the image. The following levels classify by internal images. These classifiers use artificial neural networks. The efficiency of the system on the GTSRB collection is 98 %.

Applying machine learning methods to predict geology using soil sample geochemistry

In this study we compared various machine learning techniques that used soil geochemistry to aid in geologic mapping. We tested six different sampling methods (undersample, oversample, Synthetic Minority Oversampling Technique (SMOTE), Adaptive Synthetic Sampling (ADASYN), SMOTE and Edited Nearest Neighbor (SMOTEENN), and SMOTE and Tomek links (SMOTETomek)). SMOTE performed best with ADASYN and SMOTETomek having slightly lower effectiveness. Nine machine learning algorithms (naïve Bayes, logistic regression, quadratic discriminant analysis, nearest neighbors, radial basis function support-vector machine, artificial neural network, random forest, AdaBoost classifier, and gradient boosting classifier) were compared and AdaBoost classifiers and gradient boosting classifiers were found to be most effective. Finally, we experimented with multiple classifier systems (MCS) testing different combinations of algorithms and various combinatorial functions. It was found that MCS can outperform individual models, and the best MCS combined nearest neighbors, radial basis function support-vector machine, artificial neural network, random forest, AdaBoost classifiers, and gradient boosting classifier, then applied a logistic regression to the probabilities output by the models. Ultimately, we created a tool that is able to adequately predict underlying geology in the study area using soil geochemistry.

Remote Drowsiness Detection Based on the mmWave FMCW Radar

Drowsiness can lead to inefficiency and major disasters; thus it is important to address it in both academia and industry. Despite multiple types of research in this field, a nonintrusive classifier system for detecting drowsiness in real-time under a natural environment without specific stimulation is lacking. This study develops a real-time drowsiness detection system using a 77 GHz millimeter-wave (mmWave) frequency-modulated continuous wave radar. Specifically, firstly, a real-time mmWave processing module is proposed, which can adaptively suppress both stationary and non-stationary clutters. Secondly, a feature extraction module based on a hybrid of handcrafted and machine learning (ML) features is propsoed to obtain a holistic view of mmWave-based vital signals, in which ML features represent linear and temporal changes. Thirdly, a drowsiness classification model is proposed based on feature fusion and extreme gradient boosting algorithms, thus classifying a user’s state into two categories: non-drowsy and drowsy. To confirm the proposed system’s performance, it is validated on a self-collected dataset (n = 28). The experimental results show the following (i) the accuracy of heart rate evaluation is 96.4%, and (ii) based on 10-fold cross-validation, the proposed system gains a detection accuracy of 82.9% and outperforms the state-of-the-art approaches.

A multiple classifiers system with roulette-based feature subspace selection for one-vs-one scheme

A multiple classifiers system with roulette-based feature subspace selection for one-vs-one scheme

Data-driven learning fatigue detection system: A multimodal fusion approach of ECG (electrocardiogram) and video signals

Fatigue could lead to low efficiency and even serious disaster. In the educational field, detecting fatigue could help adjust teaching strategies accordingly when a student is inactive, which can potentially improve learning efficiency. Despite numerous studies in fatigue detection, there is still a lack of multiple classifier systems capable of detecting fatigue in daily life (without specific stimulations). To initially alleviate this problem, this study develops a learning fatigue detection system using a multimodal approach with ECG and video signals, classifying a learner’s state into three categories: alert, normal, and fatigued. To validate performance, the proposed system is tested on (i) an open-source dataset DROZY (n = 35) and (ii) a self-collected dataset captured in a learning environment (n = 92). The experimental results based on 10-fold cross-validation demonstrate that the system outperforms the state-of-the-art approaches, achieving a detection accuracy of 99.6% and 91.8% on the two datasets, respectively.

Homogenous Multiple Classifier System for Software Quality Assessment Based on Support Vector Machine

In today's society, almost all human endeavours depend on software products. Lack of quality software is one of the software industry's most important problems. Hence, it would be beneficial to access the quality of software to improve and enhance software products while increasing customer satisfaction. This paper assesses software product quality using a Support Vector Machine-based ensemble classifier. The ISO/IEC-9126 (International Organization for Standardization 2001) software quality (SQ) framework was adopted in this work. Dimension reduction of the product metric category dataset and the entire PM dataset was conducted using linear discriminant analysis (LDA). SVM kernel functions (linear, quadratic, cubic, fine gaussian, medium gaussian and coarse gaussian) were used to model each classifier. The combinations of the results from the multiple SVMs used AdaBoost, bagging, and random subspace ensemble methods for the assessment of SQ. All three ensemble learning methods performed better than the individual SVM, however, the bagging stood out with an accuracy of 93.0%. Hence, it was adopted in the fusion of the SVM results and classification of SQ into classes. Results from the confusion matrix and receivers’ operating characteristics were greater than 97.99% and confirm significant improvements with an ensemble of homogenous classifiers based on SVM.