Learning Vector Quantization Model Research Articles

Low-Power Analog Integrated Architecture of the Voting Classification Algorithm for Diabetes Disease Prediction.

A low-power (∼ 600nW), fully analog integrated architecture for a voting classification algorithm is introduced. It can effectively handle multiple-input features, maintaining exceptional levels of accuracy and with very low power consumption. The proposed architecture is based on a versatile Voting algorithm that selectively incorporates one of three key classification models: Bayes or Centroid, or, the Learning Vector Quantization model; all of which are implemented using Gaussian-likelihood and Euclidean distance function circuits, as well as a current comparison circuit. To evaluate the proposed architecture, a comprehensive comparison with popular analog classifiers is performed, using real-life diabetes dataset. All model architectures were trained using Python and compared with the software-based classifiers. The circuit implementations were performed using the TSMC 90 nm CMOS process technology and the Cadence IC Suite was utilized for the design, schematic and post-layout simulations. The proposed classifiers achieved sensitivity of ≥ 96.7% and specificity of ≥ 89.7%.

Classification of Palm Fruit Ripeness Level Using Learning Vector Quantization (LVQ) Method

The implementation of Learning Vector Quantization (LVQ) to classify the ripeness of oil palm fruits is investigated in this study. In addition, this study provides a comprehensive set of procedures ranging from data collection and pre-processing to training and testing of the LVQ model, and finally, the proposed method has been validated by testing it on previously unseen data. Three feature extraction methods, specifically Gray-Level Co-occurrence Matrix (GLCM), Hue, Saturation, and Value (HSV), and t-Distributed Stochastic Neighbor Embedding (t-SNE), were assessed for their performance. The results show that the chosen feature extraction method strongly influences the classification performance. The accuracy of the model employing t-SNE features is notably the highest at 50%, indicating its efficacy in identifying the ripeness level of palm fruits by extracting pertinent features. On the other hand, the GLCM feature has a 40% accuracy in the test data, suggesting that although it captures information on texture, it may not comprehensively encapsulate ripeness characteristics. Additionally, the HSV feature achieves an accuracy of 45%, which is less precise than that of t-SNE. To conclude, this study elucidates the appropriateness of various feature extraction techniques in categorizing the degree of ripeness in palm fruits. The t-SNE feature extraction model stands out as the most efficient option, exhibiting greater precision in comparison to other methodologies.

Learning Vector Quantization based predictor model selection for hourly load demand forecasting

The development of smart grids has enabled a wide variety of generation units to be included in the grid, which has made it necessary to predict the load demand to manage the grid accurately. In this study, a novel hybridization approach is proposed based on selecting the most appropriate method to be used in each prediction step. In this scope, Learning Vector Quantization (LVQ) is employed as a classifier while Elman Neural Networks (ENN), and Ridge Regression are selected as predictors. In the first stage, a forecasting model is built and trained with ENN and Ridge Regression models on training data using 1-h before, 2-h before, and the first-order derivative of electricity consumption data. In the second stage, the method by which the most successful results are obtained for each forecast is determined and labeled. Then the LVQ model is built and trained to determine the most accurate forecast by using the same input employed in modeling. Finally, forecasts are performed by deciding which model to be used with LVQ. Experimental results show that more accurate forecasting performance is obtained with the proposed approach than the other two individual models. The different combinations of conventional models are used to illustrate the effectiveness of the proposed selection strategy, and experimental results show that better forecasting performance is obtained with the proposed approach than the individual ones at each combination.

Learning vector quantization model for prediction of cerebral Aβ in non‐demented individuals using plasma biomarkers

AbstractBackgroundCerebral Aβ is the hallmark of the AD in the early therapeutic intervention as well as diagnosis of the disease. Detecting in‐vivo Aβ deposition, however, still have limited accessibility. In this respect, the current study aims to identify plasma markers related to cerebral Aβ status via simple machine learning approach.MethodsPredictability of the candidate plasma markers for cerebral Aβ burden was examined by analyzing 303 non‐demented participants those who had rules based medicine (RMB) and amyloid PET data from the ADNI cohort at baseline visit (83 control and 220 with MCI). Demographic variables (age, gender, education, and APOE status), and 146 of 190 plasma analytes passed quality check were selected as input in learning vector quantization (LVQ) model. The cerebral Aβ burden was measured using PiB PET images. Predictability and importance of variable were calculated using LVQ model.ResultsLVQ model with 10‐cross validation predicted cerebral Aβ status with the 90.67% accuracy (balanced accuracy: 93.14%) in non‐demented people. Based on variable importance analysis, Top‐10 variables above importance ratio above 60 after rescaling were selected: APOE carrier status (100.00), Angiotensin‐Converting Enzyme (ACE, 92.77), Eotaxin‐3 (89.74), Chromogranin A (CgA, 83.61), Brain Natriuretic Peptide (BNP, 79.61), Creatine Kinase‐MB (CK‐MB, 69.61), Monokine Induced by Gamma Interferon (68.26), T Lymphocyte‐Secreted Protein I‐309 (I‐309, 66.89), Matrix Metalloproteinase‐9 (MMP‐9, 64.52), Matrix Metalloproteinase‐9 total (MMP‐9 total, 60.90).ConclusionsOur findings demonstrated that plasma based analytes can predict the cerebral Aβ positivity in non‐demented people. More specifically, ACE is can be a candidate marker for early detection of Aβ deposition in the asymptomatic stage of dementia. Plus, Eoxatain, and CgA can also play a role in AD pathology. To this end, this research was supported by the MSIT (Ministry of Science and ICT), Korea, under the ITRC (Information Technology Research Center) support program(IITP‐2021‐2017‐0‐01630) supervised by the IITP (Institute for Information & communications Technology Promotion).

Efficient computation of counterfactual explanations and counterfactual metrics of prototype-based classifiers

The increasing use of machine learning in practice and legal regulations like EU’s GDPR cause the necessity to be able to explain the prediction and behavior of machine learning models. A prominent example of particularly intuitive explanations of AI models in the context of decision making are counterfactual explanations. Yet, it is still an open research problem how to efficiently compute counterfactual explanations for many models. In this contribution, we investigate how to efficiently compute counterfactual explanations for an important class of models, prototype-based classifiers such as learning vector quantization models. In particular, we derive specific convex and non-convex programs depending on the used metric. Typically counterfactual explanations deliver a feedback in terms of changes of the input features which lead to a different output – one application scenario is the link of these required changes to actionable items to change the desired outcome. Yet, rather than minimum changes of the input, it is interesting to address minimum changes of the model itself, which are required to lead to a different result. rather than a change of its inputs. We phrase this question as a counterfactual of the model prescription rather than the data points. We focus on distance-based classifiers (in particular learning vector quantization models), where model changes correspond to changes of metric parameters, and we develop efficient optimization techniques to generate such counterfactual metric changes depending on the chosen model.

Identification of Pilots’ Fatigue Status Based on Electrocardiogram Signals

Fatigue is an important factor affecting modern flight safety. It can easily lead to a decline in pilots’ operational ability, misjudgments, and flight illusions. Moreover, it can even trigger serious flight accidents. In this paper, a wearable wireless physiological device was used to obtain pilots’ electrocardiogram (ECG) data in a simulated flight experiment, and 1440 effective samples were determined. The Friedman test was adopted to select the characteristic indexes that reflect the fatigue state of the pilot from the time domain, frequency domain, and non-linear characteristics of the effective samples. Furthermore, the variation rules of the characteristic indexes were analyzed. Principal component analysis (PCA) was utilized to extract the features of the selected feature indexes, and the feature parameter set representing the fatigue state of the pilot was established. For the study on pilots’ fatigue state identification, the feature parameter set was used as the input of the learning vector quantization (LVQ) algorithm to train the pilots’ fatigue state identification model. Results show that the recognition accuracy of the LVQ model reached 81.94%, which is 12.84% and 9.02% higher than that of traditional back propagation neural network (BPNN) and support vector machine (SVM) model, respectively. The identification model based on the LVQ established in this paper is suitable for identifying pilots’ fatigue states. This is of great practical significance to reduce flight accidents caused by pilot fatigue, thus providing a theoretical foundation for pilot fatigue risk management and the development of intelligent aircraft autopilot systems.

Real-Time Behavioral Recognition in Dairy Cows Based on Geomagnetism and Acceleration Information

The behaviors of dairy cows, such as feeding, ruminating, running, resting (standing, lying), head-shaking, drinking, and walking, can indicate their health status. In this study, a multi-sensor was used to collect data of cow's multi-behaviors for research on behavior recognition. Firstly, a collar style data acquisition system equipped with geomagnetic and acceleration sensors to collect the behavioral data of dairy cows during their daily activities was designed. Secondly, the dairy cow behavioral recognition fusion model based on K-Nearest-Neighbors (KNN) and Random Forest (RF) models were used for behavior classification. To verify the accuracy of the fusion model, the algorithms of KNN, RF, Gradient Boosting Decision Tree (GBDT), Support Vector Machine (SVM), and Learning Vector Quantization (LVQ) were introduced for comparative recognition experiments with different algorithms. The KNN-RF fusion model had the highest average recognition accuracy of 98.51%, followed by the KNN model with an average recognition accuracy of 95.37%, and the LVQ model had the lowest average recognition accuracy of 80.81%. For the recognition and verification of each behavior, the KNN-RF fusion model had the most obvious improvement in the recognition of dairy cow feeding behavior, with a recognition accuracy of 99.34%, followed by the KNN model with a recognition accuracy of 95.07%. All six models had the lowest recognition accuracy for cow head-shaking behavior: a recognition accuracy of 89.11% with the KNN-RF model followed by the RF model with a recognition accuracy of 85.14%. The system can quickly and continuously collect cow behavior information, accurately recognize individual behaviors, and provide a scientific basis for the optimal design and efficient management of digital facilities and equipment for dairy cows.

Comparison of error and uncertainty of decision tree and learning vector quantization models for predicting soil classes in areas with low altitude variations

Digital soil maps illustrate the spatial distribution of soil classes or properties and document the error and uncertainty of the soil class prediction. We assessed the potential of the decision tree (DT) and learning vector quantization (LVQ) models for prediction of soil classes in the Shahrekord plain (with low altitude variations), Iran, at different levels of Soil Taxonomy (ST) and World Reference Base (WRB) classification systems, and analyzed the error and uncertainty of both models. Two comprehensive datasets were used to predict the soil classes including soil characteristics derived from 120 excavated pedons using a stratified sampling scheme in the study area, and some auxiliary parameters (such as covariates of a digital elevation model). The cross-validation method was used to determine the uncertainty of the models. Results showed that the error and uncertainty of soil class prediction increased from the high levels towards lower levels in both soil classification systems. The first and second levels of the WRB system correlated with the suborder and subgroup levels of the ST system, respectively, which was also reflected in similar errors of these models for the predicted soil classes. The error and uncertainty in the LVQ model was remarkably higher than those of the DT model, proposing a higher accuracy of the DT model for prediction of soil classes in areas with low altitude variations. However, the LVQ model was demonstrated to be a more reliable model where the number of soil classes is low.

Expectation maximization transfer learning and its application for bionic hand prostheses

Machine learning models in practical settings are typically confronted with changes to the distribution of the incoming data. Such changes can severely affect the model performance, leading for example to misclassifications of data. This is particularly apparent in the domain of bionic hand prostheses, where machine learning models promise faster and more intuitive user interfaces, but are hindered by their lack of robustness to everyday disturbances, such as electrode shifts. One way to address changes in the data distribution is transfer learning, that is, to transfer the disturbed data to a space where the original model is applicable again. In this contribution, we propose a novel expectation maximization algorithm to learn linear transformations that maximize the likelihood of disturbed data after the transformation. We also show that this approach generalizes to discriminative models, in particular learning vector quantization models. In our evaluation on data from the bionic prostheses domain we demonstrate that our approach can learn a transformation which improves classification accuracy significantly and outperforms all tested baselines, if few data or few classes are available in the target domain.

An Influence of Measurement Scale of Predictor Variable on Logistic Regression Modeling and Learning Vector Quntization Modeling for Object Classification

Much real world decision making is based on binary categories of information that agree or disagree, accept or reject, succeed or fail and so on. Information of this category is the output of a classification method that is the domain of statistical field studies (eg Logistic Regression method) and machine learning (eg Learning Vector Quantization (LVQ)). The input argument of a classification method has a very crucial role to the resulting output condition. This paper investigated the influence of various types of input data measurement (interval, ratio, and nominal) to the performance of logistic regression method and LVQ in classifying an object. Logistic regression modeling is done in several stages until a model that meets the suitability model test is obtained. Modeling on LVQ was tested on several codebook sizes and selected the most optimal LVQ model. The best model of each method compared to its performance on object classification based on Hit Ratio indicator. In logistic regression model obtained 2 models that meet the model suitability test is a model with predictive variables scaled interval and nominal, while in LVQ modeling obtained 3 pieces of the most optimal model with a different codebook. In the data with interval-scale predictor variable, the performance of both methods is the same. The performance of both models is just as bad when the data have the predictor variables of the nominal scale. In the data with predictor variable has ratio scale, the LVQ method able to produce moderate enough performance, while on logistic regression modeling is not obtained the model that meet model suitability test. Thus if the input dataset has interval or ratio-scale predictor variables than it is preferable to use the LVQ method for modeling the object classification.

Median variants of learning vector quantization for learning of dissimilarity data

Exemplar based techniques such as affinity propagation represent data in terms of typical exemplars. This has two benefits: (i) the resulting models are directly interpretable by humans since representative exemplars can be inspected in the same way as data points, (ii) the model can be applied to any dissimilarity measure including non-Euclidean or non-metric settings. Most exemplar based techniques have been proposed in the unsupervised setting only, such that their performance in supervised learning tasks can be weak depending on the given data. We address the problem of learning exemplar-based models for general dissimilarity data in a discriminative framework in this contribution. For this purpose, we consider variants of Kohonen׳s learning vector quantization model to handle data with only dissimilarities between available. Here the exemplars are the prototypes. The resulting classifiers represent data in terms of sparse models thereby reaching state-of-the art results in benchmarks. For a real world data set in the field of veterinary medicine we report promising results.

Classification of Postural Profiles among Mouth-breathing Children by Learning Vector Quantization

Mouth breathing is a chronic syndrome that may bring about postural changes. Finding characteristic patterns of changes occurring in the complex musculoskeletal system of mouth-breathing children has been a challenge. Learning vector quantization (LVQ) is an artificial neural network model that can be applied for this purpose. The aim of the present study was to apply LVQ to determine the characteristic postural profiles shown by mouth-breathing children, in order to further understand abnormal posture among mouth breathers. Postural training data on 52 children (30 mouth breathers and 22 nose breathers) and postural validation data on 32 children (22 mouth breathers and 10 nose breathers) were used. The performance of LVQ and other classification models was compared in relation to self-organizing maps, back-propagation applied to multilayer perceptrons, Bayesian networks, naive Bayes, J48 decision trees, k, and k-nearest-neighbor classifiers. Classifier accuracy was assessed by means of leave-one-out cross-validation, area under ROC curve (AUC), and inter-rater agreement (Kappa statistics). By using the LVQ model, five postural profiles for mouth-breathing children could be determined. LVQ showed satisfactory results for mouth-breathing and nose-breathing classification: sensitivity and specificity rates of 0.90 and 0.95, respectively, when using the training dataset, and 0.95 and 0.90, respectively, when using the validation dataset. The five postural profiles for mouth-breathing children suggested by LVQ were incorporated into application software for classifying the severity of mouth breathers' abnormal posture.

Noise Fuzzy Learning Vector Quantization

Fuzzy learning vector quantization (FLVQ) benefits from using the membership values coming from fuzzy c-means (FCM) as learning rates and it overcomes several problems of learning vector quantization (LVQ). However, FLVQ is sensitive to noises because it is a FCM-based algorithm (FCM is sensitive to noises). Here, a new fuzzy learning vector quantization model, called noise fuzzy learning vector quantization (NFLVQ), is proposed to handle the noises sensitivity problem of FLVQ. NFLVQ integrates LVQ and generalized noise clustering (GNC), uses the membership values from GNC as learning rates and clusters data containing noisy data better than FLVQ. Experimental results show the better performances of NFLVQ.

Predicting the Androgenicity of Structurally Diverse Compounds from Molecular Structure Using Different Classifiers

AbstractMany environmental and industrial chemicals are reported to have androgenic or antiandrogenic activities. These androgenic chemicals may act as hormones and have the potential to disrupt the endocrine systems of wildlife and humans. In this study, the probabilistic neural network (PNN), support vector machine (SVM), and learning vector quantization (LVQ), three types of machine learning, were used to develop binary classification models to predict androgenicity directly from the organic compounds' molecular structures which were represented by only eleven numerical descriptors. The PNN model acquired the best overall classification rate of 86.67% for prediction data set, with Matthews Correlation Coefficient of 0.64, and the LVQ model gave the lowest false negative rate of 0.00%, which will tend to give relatively high priority during toxicology evaluation. In addition, a consensus model was produced that integrated all three of the basic model types. Compared with the individual models, this consensus model correctly predicted the androgenicity of 86.67% of the prediction set compounds, with false negative rate of 0.00% and the highest Matthews Correlation Coefficient of 0.65. The obtained results indicate that the proposed classification models could provide a feasible and practical tool for the rapid screening of potential androgens.

The learning vector quantization algorithm applied to automatic text classification tasks

Automatic text classification is an important task for many natural language processing applications. This paper presents a neural approach to develop a text classifier based on the Learning Vector Quantization (LVQ) algorithm. The LVQ model is a classification method that uses a competitive supervised learning algorithm. The proposed method has been applied to two specific tasks: text categorization and word sense disambiguation. Experiments were carried out using the Reuters-21578 text collection (for text categorization) and the Senseval-3 corpus (for word sense disambiguation). The results obtained are very promising and show that our neural approach based on the LVQ algorithm is an alternative to other classification systems.

Invariant Gender Identification

Tolba, A. S., Invariant Gender Identification, Digital Signal Processing 11 (2001) 222–240 In this paper, we address the problem of gender identification using different neural network classifiers: a learning vector quantization (LVQ) network and a radial basis function (RBF) network. Our results indicate that it is more favorable to use either the LVQ network or the RBF network than any feature-based methods. We present results showing identification of gender with a hit rate of 100% in the case of a LVQ network and 98.04% in the case of an RBF network. When hair information was excluded, the best LVQ classifier resulted in 95.1% correct identification. We show that while the two models are nearly accurate, the RBF model learns the task considerably faster than the LVQ model. These results are favorable compared with eigen-decomposition-based techniques. The effect of head covers (e.g., a scarf) used by both men and women on system performance is studied.

Computer-aided diagnosis for surgical office-based breast ultrasound.

The computer-aided diagnostic system is an intelligent system with great potential for categorizing solid breast nodules. It can be used conveniently for surgical office-based digital ultrasonography (US) of the breast. Retrospective, nonrandomized study. University teaching hospital. We retrospectively reviewed 243 medical records of digital US images of the breast of pathologically proved, benign breast tumors from 161 patients (ie, 136 fibroadenomas and 25 fibrocystic nodules), and carcinomas from 82 patients (ie, 73 invasive duct carcinomas, 5 invasive lobular carcinomas, and 4 intraductal carcinomas). The digital US images were consecutively recorded from January 1, 1997, to December 31, 1998. The physician selected the region of interest on the digital US image. Then a learning vector quantization model with 24 autocorrelation texture features is used to classify the tumor as benign or malignant. In the experiment, 153 cases were arbitrarily selected to be the training set of the learning vector quantization model and 90 cases were selected to evaluate the performance. One experienced radiologist who was completely blind to these cases was asked to classify these tumors in the test set. Contribution of breast US to diagnosis. The performance comparison results illustrated the following: accuracy, 90%; sensitivity, 96.67%; specificity, 86.67%; positive predictive value, 78.38%; and negative predictive value, 98.11% for the computer-aided diagnostic (CAD) system and accuracy, 86.67%; sensitivity, 86.67%; specificity, 86.67%; positive predictive value, 76.47%; and negative predictive value, 92.86% for the radiologist. The proposed CAD system provides an immediate second opinion. An accurate preoperative diagnosis can be routinely established for surgical office-based digital US of the breast. The diagnostic rate was even better than the results of an experienced radiologist. The high negative predictive rate by the CAD system can avert benign biopsies. It can be easily implemented on existing commercial diagnostic digital US machines. For most available diagnostic digital US machines, all that would be required for the CAD system is only a personal computer loaded with CAD software.

Neural network architectures for industrial applications

An overview of the present industrial scenario with regard to the application of neural network approaches is reported. A brief summary of practical applications is presented. Neural network architectures capable of resolving various types of industrial problems are also illustrated. The first application presented deals with the problem of digit recognition. The proposed architecture is composed of different multilayer networks trained by back-propagation, organized in a hierarchical structure. Such a structure as opposed to a single network structure, allows the avoidance of typical errors occurring on comparison of certain couples of digits. The second application deals with a texture-based segmentation of an image. In order to solve this problem, a structure composed of two different networks is proposed. The first network, a simplified BCS model, is a “feature extractor”; the second network, Learning Vector Quantization model, is able to classify different textures on the basis of the parameters extracted by BCS. The third application deals with a fault-diagnosis system for the process state monitoring of complex industrial plants. An innovative approach is presented. The proposed model analyzes process parameters and predicts possible malfunctions on the basis of self-organizing networks.