Performance Of Linear Discriminant Analysis Research Articles

Comparison of the performance of linear discriminant analysis and binary logistic regression applied to risk factors for mortality in Ebola virus disease patients

Our study aimed to identify risk factors associated with mortality in Ebola patients using binary logistic regression analysis and linear discriminant analysis, and to assess the predictive power of these two methods. Our study was a randomized, double-blind, controlled (observational) clinical trial conducted in 2018 during the 10th Ebola outbreak in eastern DRC. The study included 363 patients divided into two treatment arms, including 182 patients treated with MAB114 (Ebang) and 181 patients treated with REGENERON (REGN-EB3 ). After a thorough analysis of the data, both statistical analysis methods selected the same set of variables (risk factors) for the binary logistic regression we obtained: viral load 0.58 (0.5-0.67), creatinine 1.98 (1.58-0.67) and aspartate aminotransferase 0.99 (0.9-1); as for the linear discriminant analysis we have viral load (0.88), creatinine (0.94) and aspartate aminotransferase (0.78). We also see almost the same results when different prediction probabilities are evaluated. Logistic regression predicted a mortality rate of 36.5% and linear discriminant analysis predicted a mortality rate of 38.8%. Using the AUC (area under the curve) score, we were able to evaluate two methods and obtain a score of 0.935 for the binary logistic regression and 0.932 for the linear discriminant analysis. According to the evaluation hypothesis, both methods give the same risk factors (viral load (Ctnp), creatinine and alanine aminotransferase (ALT)) with a probability of 93%.

Weight Vector Tuning and Asymptotic Analysis of Binary Linear Classifiers

Unlike its intercept, a linear classifier's weight vector cannot be tuned by a simple grid search. Hence, this paper proposes weight vector tuning of a generic binary linear classifier through the parameterization of a decomposition of the discriminant by a scalar which controls the trade-off between conflicting informative and noisy terms. By varying this parameter, the original weight vector is modified in a meaningful way. Applying this method to a number of linear classifiers under a variety of data dimensionality and sample size settings reveals that the classification performance loss due to non-optimal native hyperparameters can be compensated for by weight vector tuning. This yields computational savings as the proposed tuning method reduces to tuning a scalar compared to tuning the native hyperparameter, which may involve repeated weight vector generation along with its burden of optimization, dimensionality reduction, etc., depending on the classifier. It is also found that weight vector tuning significantly improves the performance of Linear Discriminant Analysis (LDA) under high estimation noise. Proceeding from this second finding, an asymptotic study of the misclassification probability of the parameterized LDA classifier in the growth regime where the data dimensionality and sample size are comparable is conducted. Using random matrix theory, the misclassification probability is shown to converge to a quantity that is a function of the true statistics of the data. Additionally, an estimator of the misclassification probability is derived. Finally, computationally efficient tuning of the parameter using this estimator is demonstrated on real data.

Novel colorimetric sensor array for Chinese rice wine evaluation based on color reactions of flavor compounds

AbstractIn this study, a 3 × 3 comprehensive colorimetric sensor array (CSA) was constructed for the quality evaluation of Chinese rice wine based on three different action principles (pH change principle, indicators displacement assay principle, and ethanol fading principle). First, China national standard was used to determine the total acid, reducing sugar, and alcoholic strength of six kinds of rice wine, which forms the basis for the selection of the CSA sensitive points. Then, CSA was combined with pattern recognition technologies to qualitatively distinguish rice wine. The results showed that the when PCs = 3, the linear discriminant analysis model can achieve a recognition rate of 100%, which shows strong prediction ability compared with K‐nearest neighbor and back‐propagation artificial neural network models. Finally, CSA was combined with partial least squares (PLS) and support vector regression (SVR) to quantitatively analyze the three components in rice wine. The correlation coefficients of the training set and predictions are more than 85% in the SVR model, which is superior to the PLS model. Based on these findings, it is possible to conclude that the CSA can be used to rapidly evaluate the quality of rice wine.Practical Applications Established the 3 × 3 colorimetric sensor array (CSA) for the quality evaluation of Chinese rice wine based on three different action principles (pH change principle, indicators displacement assay principle, and ethanol fading principle). The CSA was combined with pattern recognition technologies to qualitatively distinguish different kinds of rice wine. The performance of linear discriminant analysis shows strong prediction ability compared with K‐nearest neighbor and back‐propagation artificial neural network. The CSA was combined with partial least squares (PLS) and support vector regression (SVR) to quantitatively analyze the three components in rice wine. The correlation coefficients of the training set and predictions are more than 85% in the SVR model, which is superior to the PLS model.

Intention Prediction and Human Health Condition Detection in Reaching Tasks with Machine Learning Techniques.

Detecting human motion and predicting human intentions by analyzing body signals are challenging but fundamental steps for the implementation of applications presenting human–robot interaction in different contexts, such as robotic rehabilitation in clinical environments, or collaborative robots in industrial fields. Machine learning techniques (MLT) can face the limit of small data amounts, typical of this kind of applications. This paper studies the illustrative case of the reaching movement in 10 healthy subjects and 21 post-stroke patients, comparing the performance of linear discriminant analysis (LDA) and random forest (RF) in: (i) predicting the subject’s intention of moving towards a specific direction among a set of possible choices, (ii) detecting if the subject is moving according to a healthy or pathological pattern, and in the case of discriminating the damage location (left or right hemisphere). Data were captured with wearable electromagnetic sensors, and a sub-section of the acquired signals was required for the analyses. The possibility of detecting with which arm (left or right hand) the motion was performed, and the sensitivity of the MLT to variations in the length of the signal sub-section were also evaluated. LDA and RF prediction accuracies were compared: Accuracy improves when only healthy subjects or longer signals portions are considered up to 11% and at least 10%, respectively. RF reveals better estimation performance both as intention predictor (on average 59.91% versus the 62.19% of LDA), and health condition detector (over 90% in all the tests).

Distinguish Fritillaria cirrhosa and non-Fritillaria cirrhosa using laser-induced breakdown spectroscopy

As traditional Chinese medicines, Fritillaria from different origins are very similar and it is difficult to distinguish them. In this study, the laser-induced breakdown spectroscopy combined with learning vector quantization (LIBS-LVQ) was proposed to distinguish the powdered samples of Fritillaria cirrhosa and non-Fritillaria cirrhosa. We also studied the performance of linear discriminant analysis, and support vector machine on the same data set. Among these three classifiers, LVQ had the highest correct classification rate of 99.17%. The experimental results demonstrated that the LIBS-LVQ model could be used to differentiate the powdered samples of Fritillaria cirrhosa and non-Fritillaria cirrhosa.

Application of Linear Discriminant Model to Evaluate the Association Between Milk Production, Reproductive Performance, and Calving Season in Dairy Cattle

This study was aimed to determine the performance of linear discriminant analysis (LDA) for differentiation between seasons of calving on the basis of305day- milk yield, fat %, protein %, days open (DO), days to first insemination (DFI), and number of services per conception. By considering the assumption of this method, a random sample was selected from the animals being represented by all explanatory variables. The discrimination between seasons of calving was depended on the significance of coefficients, classification rate, in addition to the group centroids. Results showed that LDA method selected 305day-milk yield (kg), days open (DO), days to first insemination and number of services per conception, as the significant (P < 0.05) contributors for data classification. The total variance explained by 2 functions was 79.2% and 15.9%, respectively. So, the 1st function can do well in discrimination process than the 2ndone. The percentage of correct classification was 64.6%. In conclusion, LDA can be used effectively for classification of calving seasons, even with violation of normality assumption.

Study of Linear Discriminant Analysis to Identify Baby Cry Based on DWT and MFCC

Baby crying is a common behavior among babies and is a means of verbal communication for babies who express their needs and desires. A baby’s cry identification system is needed because it makes it easy for adults to find out the meaning of a baby’s cry. This study proposes a system for classifying infant crying sounds using Linear Discriminant Analysis (LDA) with Discrete Wavelet Transform (DWT) and Mel-frequency Cepstral Coefficient (MFCC) as a feature extraction method. Based on experiments, the system can identify the sound of crying babies grouped into 5 (five) classes, namely discomfort, hunger, colds, burp, and drowsiness. The system achieves an accuracy of 94% and an average computing time of 1.5506 seconds. The performance of Linear Discriminant Analysis (LDA) outperformed Principal Component Analysis (LDA) in the identification of crying babies

CRLEDD: Regularized Causalities Learning for Early Detection of Diseases Using Electronic Health Record (EHR) Data

The availability of Electronic Health Records (EHR) in health care settings has provided tremendous opportunities for early disease detection. While many supervised learning models have been adopted for EHR-based disease early detection, the ill-posed inverse problem in the parameter learning has imposed a significant challenge on improving the accuracy of these algorithms. In this paper, we propose CRLEDD - Causality-Regularized Learning for Early Detection of Disease, an algorithm to improve the performance of Linear Discriminant Analysis (LDA) on top of diagnosis-frequency vector data representation. While most existing regularization methods exploit sparsity regularization to improve detection performance, CRLEDD provides a unique perspective by ensuring positive semi-definiteness of the sparsified precision matrix used in LDA which is different from the regular regularization method (e.g., L2 regularization). To achieve this goal, CRLEDD employs Graphical Lasso to estimate the precision matrix in the ill-posed settings for enhanced accuracy of LDA classifiers. We perform extensive evaluation of CRLEDD using a large-scale real-world EHR dataset to predict mental health disorders (e.g., depression and anxiety) of college students from 10 universities in the U.S. We compare CRLEDD with other regularized LDA and downstream classifiers. The result shows that CRLEDD outperforms all baselines in terms of accuracy and F1 scores.

Evaluation of surface EMG-based recognition algorithms for decoding hand movements

Myoelectric pattern recognition (MPR) to decode limb movements is an important advancement regarding the control of powered prostheses. However, this technology is not yet in wide clinical use. Improvements in MPR could potentially increase the functionality of powered prostheses. To this purpose, offline accuracy and processing time were measured over 44 features using six classifiers with the aim of determining new configurations of features and classifiers to improve the accuracy and response time of prosthetics control. An efficient feature set (FS: waveform length, correlation coefficient, Hjorth Parameters) was found to improve the motion recognition accuracy. Using the proposed FS significantly increased the performance of linear discriminant analysis, K-nearest neighbor, maximum likelihood estimation (MLE), and support vector machine by 5.5%, 5.7%, 6.3%, and 6.2%, respectively, when compared with the Hudgins’ set. Using the FS with MLE provided the largest improvement in offline accuracy over the Hudgins feature set, with minimal effect on the processing time. Among the 44 features tested, logarithmic root mean square and normalized logarithmic energy yielded the highest recognition rates (above 95%). We anticipate that this work will contribute to the development of more accurate surface EMG-based motor decoding systems for the control prosthetic hands.

Rapid and nondestructive fraud detection of palm oil adulteration with Sudan dyes using portable NIR spectroscopic techniques

ABSTRACTNon-destructive, simple and fast techniques for identifying authentic palm oil and those adulterated with Sudan dyes using portable NIR spectroscopy would be very beneficial to West Africa countries and the world at large. In this study, a portable NIR spectroscopy coupled with multivariate models were developed for detecting palm oil adulteration. A total of 520 samples of palm oil were used comprising; 40 authentic samples together with 480 adulterated samples containing Sudan dyes (I, II, III, IV of 120 samples each). Multiplicative scatter correction (MSC) preprocessing technique plus Principal component analysis (PCA) was used to extract relevant spectral information which gave visible cluster trends for authentic samples and adulterated ones. The performance of Linear discriminant analysis (LDA) and Support vector machine (SVM) were compared, and SVM showed superiority over LDA. The optimised results by cross-validation revealed that MSC-PCA + SVM gave an identification rate above 95% for both calibration and prediction sets. The overall results show that portable NIR spectroscopy together with MSC-PCA + SVM model could be used successfully to identify authentic palm oils from adulterated ones. This would be useful for quality control officers and consumers to manage and control Sudan dyes adulteration in red palm oil.

Performance evaluation of various classifiers for predicting knee angle from electromyography signals

AbstractThis paper proposes a classification‐based knee angle prediction from myoelectric signals. Surface electromyographic signals were recorded from four muscles in the lower limb while performing the task of standing up from a chair and sitting down on the chair. Knee angle was measured using a goniometer and quantised into five levels/classes. The surface electromyographic signals were segmented using overlapped windowing. Fifteen features per muscle were extracted and fed to the classifier. The classifier predicts the class of the knee angle at a particular instant This study examines the performance of linear discriminant analysis, Naive Bayes, k‐nearest neighbour, and support vector machine classifiers. The support vector machine classifier with a quadratic kernel performed best, with a classification accuracy of 92.2 ± 2.2%, a sensitivity of 90.19 ± 3.06%, a specificity of 98.11 ± 0.63%, and 89.38 ± 3.0% precision.

Discriminant Analysis via Joint Euler Transform and ℓ2,1-norm.

Linear Discriminant analysis (LDA) has been widely used for face recognition. However, when identifying faces in the wild, the existence of outliers that deviate significantly from the rest of data can arbitrarily skew the desired solution. This usually deteriorates LDA's performance dramatically, thus preventing it from mass deployment in real-world applications. To handle this problem, we propose an effective distance metric learning method based LDA, namely Euler LDA-L21 (e-LDA-L21). e-LDA-L21 is carried out in two stages, in which each image is mapped into a complex space by Euler transform in the first stage and the ℓ2,1-norm is adopted as the distance metric in the second stage. This not only reveals nonlinear features but also exploits the geometric structure of data. To solve e-LDA-L21 efficiently, we propose an iterative algorithm, which is a closed-form solution at each iteration with convergence guaranteed. Finally, we extend e-LDA-L21 to Euler 2DLDA-L21 (e-2DLDA-L21) which further exploits the spatial information embedded in image pixels. Experimental results on several face databases demonstrate its superiority over the state-of-the-art algorithms.

Use of Quantitative Morphological and Functional Features for Assessment of Axillary Lymph Node in Breast Dynamic Contrast-Enhanced Magnetic Resonance Imaging.

Background Axillary lymph-node assessment is considered one of the most important prognostic factors concerning breast cancer survival. Objective We investigated the discriminative power of morphological and functional features in assessing the axillary lymph node. Methods We retrospectively analysed data from 52 consecutive patients who undergone DCE-MRI and were diagnosed with primary breast carcinoma: 94 lymph nodes were identified. Per each lymph node, we extracted morphological features: circularity, compactness, convexity, curvature, elongation, diameter, eccentricity, irregularity, radial length, entropy, rectangularity, roughness, smoothness, sphericity, spiculation, surface, and volume. Moreover, we extracted functional features: time to peak (TTP), maximum signal difference (MSD), wash-in intercept (WII), wash-out intercept (WOI), wash-in slope (WIS), wash-out slope (WOS), area under gadolinium curve (AUGC), area under wash-in (AUWI), and area under wash-out (AUWO). Selection of important features in predicting metastasis has been done by means of receiver operating characteristic (ROC) analysis. Performance of linear discriminant analysis was analysed. Results All morphological features but circularity showed a significant difference between median values of metastatic lymph nodes group and nonmetastatic lymph nodes group. All dynamic parameters except for MSD and WOS showed a statistically significant difference between median values of metastatic lymph nodes group and nonmetastatic lymph nodes group. Best results for discrimination of metastatic and nonmetastatic lymph nodes were obtained by AUGC (accuracy 75.8%), WIS (accuracy 71.0%), WOS (accuracy 71.0%), and AUCWO (accuracy 72.6%) for dynamic features and by compactness (accuracy 82.3%), curvature (accuracy 71.0%), radial length (accuracy 71.0%), roughness (accuracy 74.2%), smoothness (accuracy 77.2%), and speculation (accuracy 72.6%) for morphological features. Linear combination of all morphological and/or of all dynamic features did not increase accuracy in metastatic lymph nodes discrimination. Conclusions Compactness as morphological feature and area under time-intensity curve as dynamic feature were the best parameters in identifying metastatic lymph nodes on breast MRI.

Financial Distress Prediction using Linear Discriminant Analysis and Support Vector Machine

A financial difficulty is the early stages before the bankruptcy. Bankruptcies caused by the financial distress can be seen from the financial statements of the company. The ability to predict financial distress became an important research topic because it can provide early warning for the company. In addition, predicting financial distress is also beneficial for investors and creditors. This research will be made the prediction model of financial distress at industrial companies in Indonesia by comparing the performance of Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) combined with variable selection technique. The result of this research is prediction model based on hybrid Stepwise-SVM obtains better balance among fitting ability, generalization ability and model stability than the other models.

Evaluation of linear discriminant and support vector machine classifiers for classification of nitrogen status in mature oil palm from SPOT-6 satellite images: analysis of raw spectral bands and spectral indices

Nitrogen (N) management is important in sustaining oil palm production. Remote sensing-based approaches via spectral index have promise in assessing the N nutrition content. The objectives of this study are; (i) to examine the N classification capability of three spectral indices (SI) such as visible (Vis), near infrared (NIR) and a combination of visible and NIR (Vis + NIR) from the SPOT-6 satellite, and (ii) to compare the performance of linear discriminant analysis (LDA) and support vector machine (SVM) in discriminating foliar N content of mature oil palms. Nitrogen treatments varied from 0 to 2 kg per palm. The N-sensitive SIs tested in this study were age-dependent. The Vis index (BGRI1) (CVA = 79.55%) and Vis + NIR index (NDVI, NG, IPVI and GNDVI) (CVA = 81.82%) were the best indices to assess N status of young and prime mature palms through the SVM classifier.

Building Types’ Classification Using Shape-Based Features and Linear Discriminant Functions

In this paper, the applicability and performance of linear discriminant analysis (LDA) for building types’ classification are investigated. Building models at a level of detail 1 (LoD1) are derived from real estate cadastral building footprints and digital surface models from stereoscopic airborne images. In several experiments for two cities in Germany (Berlin and Munich), we first evaluate the discriminatory power of 26 different shape-based features which describe the physiognomy of individual buildings in terms of 1-D (e.g., length), 2-D (e.g., area), and 3-D (e.g., volume) features. While 1-D features show low contributions to the discrimination of the five building types, we observe high contributions of the 3-D shape index and 2-D measures of compactness. In a second group of experiments, the size of training samples for the classification process is investigated with the outcome that a size of 10% of the total number of labeled features is practicable in terms of size and accuracy. In a third battery of experiments, the selected features and training sample size are used for the classification of building types resulting in kappa values of 0.94 for both cities. In the final experiments, the geographical transfer between the two cities is investigated reaching kappa values of 0.93 and 0.91, respectively. The tests show that a simple linear classifier like LDA can handle building types’ classification without much user interaction compared to more complex classification methods but is limited when similar building types (e.g., perimeter block development and block development ) are to be discriminated.

A Comparison of Different Classification Techniques to Determine the Change Causes in Hotelling'sT2Control Chart

To detect out‐of control situations using multivariate control charts is relatively easy. However, the determination of the change causes is more difficult. In the last years the application of classification techniques to analyze the out‐of‐control signals has been proposed. These proposals include increasingly sophisticated methods but it is not clear, if there is one, which is the most powerful. In this paper we have simulated different scenarios varying the correlation structure and shift type, to test the performance of Linear Discriminant Analysis, Classification Trees, Neural Networks and Boosting Trees with the main aim to analyze whether or not using more complex classification methods is worthy. Results show that although the performance of different methods depends on the correlation structure and the class of change, under the more feasible situations Boosting shows the best performance. Copyright © 2015 John Wiley &amp; Sons, Ltd.

ROUGH SETS AND DISCRIMINANT ANALYSIS TECHNIQUES FOR BUSINESS DEFAULT FORECASTING

(ProQuest: ... denotes formulae omitted.)1. INTRODUCTIONMany studies have addressed the problem of bankruptcy prediction and most of them do their analysis through financial ratios. Beaver (1966) utilized t-tests to evaluate the predictive ability of various financial ratios using a pair-matched sample. In a landmark paper in the bankruptcy prediction area, Altman (1968) used multiple discriminant analysis on a pair-matched sample. Altman (1968) used five financial ratios to predict whether a company is a going concern. Altman's study is very accurate in its prediction of which companies are going to go bankrupt, and the study has held up very well even after almost 35 years.In another major study Ohlson (1980) applied logistic regression in a much larger sample that did not involve pair-matching, and he found that the four statistically significant factors for identifying the probability of failure are the size of the company, measures of financial structure, measures of performance, and measures of current liquidity.Lennox (1999) examined the causes of bankruptcy for a sample of UK listed companies for the period 1987 to 1994 and found that the most important determinants of bankruptcy are profitability, leverage, cash flow, company size, industry sector and the economic cycle. Lennox (1999) argues that well-specified logit and probit models can be more accurate for identifying failing companies than discriminant analysis.Barniv and McDonald (1999) discuss several alternative techniques to probit models and logit models for predicting company failure. In recent years, many studies have developed and examined new methods for predicting bankruptcy. For example, Hwang et al. (2007) and Cheng et al. (2010) use semi-parametric methods to predict bankruptcy. Serrano-Cinca and Gutierrez-Nieto (2013) use partial lease square discriminant analysis in bankruptcy prediction. In this context, Ravi Kumar and Ravi (2007) do a deep review of the papers issued on bankruptcy forecasting and classified them into two groups, according to the usage of statistical (like the papers cited up to now) or intelligent techniques. Within the latter, the authors distinguish the following techniques: neural networks, case-based reasoned techniques, decision trees, evolutionary approaches, operational research, rough sets, support vector machines, isotonic separation, fuzzy techniques and soft computing techniques.Within the group of the papers that use fuzzy techniques, Alam et al. (2000) propose fuzzy clustering for identifying potentially failed banks and then compared it with neural networks. De Andres et al. (2005) propose fuzzy rule based classifiers for bankruptcy prediction problem. They compare the performance of Linear Discriminant Analysis and logistic regression with multi-layer perception and fuzzy-rule based classifiers. Korol and Korodi (2011) propose a fuzzy model (that uses financial ratios as inputs) to forecast firms' bankruptcy up to three years before. Scherger et. al. (2014) focuses their analysis on a set of Small and Medium sized Enterprises (SMEs) from the construction sector. The authors simulate a fuzzy model to forecast firms' health and find out the reasons that generate problems (diseases) to it.Ravi Kumar and Ravi (2007) state that, among the intelligent techniques, the neural networks has been the most used family to solve the problem of bankruptcy prediction. For example, Lin and McClean (2000) conducted a study on distressed and non-distressed quoted companies listed on the London Stock Exchange for the last 20 years. Lin and McClean used different techniques in their studies such as discriminant analysis, logistic regression, neural networks and decision trees. They conclude that the latter and neural networks provide the best results. It must be underlined that, during last years, an important line of research that combines neural networks with fuzzy logic has raised. Papers like the ones of Chen et al. …

Classification of salt marsh vegetation using edaphic and remote sensing-derived variables

Salt marsh plant communities are known for their striking patterns of vertical zonation. Two of the most important edaphic parameters that affect species distribution patterns are soil salinity and waterlogging, both of which are related to topographical variations and distance to the water. The primary objective of this study was to evaluate whether information on elevation and distance derived through remote sensing could be used to predict plant distributions in a southeastern United States salt marsh. We classified four marsh vegetation classes (tall Spartina alterniflora, medium S. alterniflora/short S. alterniflora, marsh meadow, and Borrichia frutescens/Juncus roemerianus) based on landscape metrics obtained from a light detection and ranging (LIDAR)-derived digital elevation model (DEM) and compared results to a classification based on field-collected edaphic variables. Our secondary objective was to compare the performance of linear discriminant analysis (LDA) with non-parametric classification and regression trees (CART) for these classifications. Models based on the edaphic variables soil water content, salinity, and redox potential attained accuracies of 0.62 and 0.71 with LDA and CART, respectively. When the remote sensing-derived variables DEM elevation, slope, distance to the mean high water line, and distance to upland area were used, classification accuracies improved to 0.78 for LDA and 0.79 for CART. Our results suggest that remote sensing-derived metrics can capture edaphic gradients effectively, which makes them especially suited to landscape level analyses of salt marsh plant habitats, with potential application for predicting the effects of sea level rise on salt marsh plant distribution.

Analytical study of performance of linear discriminant analysis in stochastic settings

This paper provides exact analytical expressions for the first and second moments of the true error for linear discriminant analysis (LDA) when the data are univariate and taken from two stochastic Gaussian processes. The key point is that we assume a general setting in which the sample data from each class do not need to be identically distributed or independent within or between classes. We compare the true errors of designed classifiers under the typical i.i.d. model and when the data are correlated, providing exact expressions and demonstrating that, depending on the covariance structure, correlated data can result in classifiers with either greater error or less error than when training with uncorrelated data. The general theory is applied to autoregressive and moving-average models of the first order, and it is demonstrated using real genomic data.