Performance Of Different Classification Methods Research Articles

Eye Tracking-Based Stress Classification of Athletes in Virtual Reality

Monitoring stress is relevant in many areas, including sports science. In that scope, various studies showed the feasibility of stress classification using eye tracking data. In most cases, the screen-based experimental design restricted the motion of participants. Consequently, the transferability of results to dynamic sports applications remains unclear. To address this research gap, we conducted a virtual reality-based stress test consisting of a football goalkeeping scenario. We contribute by proposing a stress classification pipeline solely relying on gaze behaviour and pupil diameter metrics extracted from the recorded data. To optimize the analysis pipeline, we applied feature selection and compared the performance of different classification methods. Results show that the Random Forest classifier achieves the best performance with 87.3% accuracy, comparable to state-of-the-art approaches fusing eye tracking data and additional biosignals. Moreover, our approach outperforms existing methods exclusively relying on eye measures.

Monitoring and prediction of sensory shelf‐life in strawberry with ultraviolet‐visible‐near‐infrared (UV‐VIS‐NIR) spectroscopy

The suitability of UV-VIS-NIR reflectance for predicting the sensory shelf-life (SSL) and the number of days under refrigerated storage (DS) for strawberries was examined. The performance of different classification methods for predicting the days of storage was compared. Partial least squares regression (PLSR) models were calibrated and evaluated for predicting the number of days under storage resulting in similar performance with NIR (R2 = 0.870), UV-VIS (R2 = 0.874), and UV-VIS-NIR (R2 = 0.877) datasets in evaluation sets. The shelf-life of strawberries were estimated from the visual sensory scores from a panel. Based on the sensory shelf life of each strawberry and their days under storage, remaining days till rejection (DTR) was calculated. PLSR models were trained to predict the remaining DTR from UV-VIS-NIR reflectance datasets resulting in performance up to R2 = 0.712 in evaluation sets. The influence of the number of days of storage on the remaining DTR was examined with simple regression models. The PLSR models were trained to predict the residuals from the simple regression models to improve the prediction of the remaining DTR. The PLSR models were built from the predictor dataset containing PCA denoised and Savistzky-Golay (SG) second derivate transformed UV-VIS reflectance data and extended multiplicative signal corrected (EMSC) and first derivate SG transformed NIR reflectance data. The resulting models for predicting the remaining DTR achieved the performance up to R2 = 0.818 in evaluation sets. Repeated feature selection was carried out using Uninformative Variable Elimination (UVE-PLS) in the PLSR models predicting days of storage and remaining shelf-life. The wavebands selected more frequently by the models predicting DS and DTR with feature selection methods were recorded, visualized, and interpreted.

Assessing over Time Performance of an eNose Composed of 16 Single-Type MOX Gas Sensors Applied to Classify Two Volatiles

This paper assesses the over time performance of a custom electronic nose (eNose) composed of an array of commercial low-cost and single-type miniature metal-oxide (MOX) semiconductor gas sensors. The eNose uses 16 BME680 versatile sensor devices, each including an embedded non-selective MOX gas sensor that was originally proposed to measure the total volatile organic compounds (TVOC) in the air. This custom eNose has been used previously to detect ethanol and acetone, obtaining initial promising classification results that worsened over time because of sensor drift. The current paper assesses the over time performance of different classification methods applied to process the information gathered from the eNose. The best classification results have been obtained when applying a linear discriminant analysis (LDA) to the normalized conductance of the sensing layer of the 16 MOX gas sensors available in the eNose. The LDA procedure by itself has reduced the influence of drift in the classification performance of this single-type eNose during an evaluation period of three months.

Global spatiotemporal consistency between meteorological and soil moisture drought indices

In this study, the consistency of a set of meteorological and soil moisture drought indices has been analyzed using the linear relationship between them. Commonly used Standardized Precipitation Index (SPI) and Standardized Precipitation Evapotranspiration Index (SPEI) are compared to soil moisture drought index (SSMI) between the years 1981 and 2019. The most consistent meteorological drought index with SSMI has been selected as the best representative of it for 175,840 pixels, globally. Later, different classification methods (i.e., Decision Tree, K-Nearest Neighbors, Naïve Bayesian, Support Vector Machine, and Random Forest) are trained to identify the best representative of SSMI by considering the most consistent meteorological drought index with SSMI as target, and five ancillary datasets of average precipitation, average temperature, vegetation cover (Normalized difference vegetation index; NDVI), climate class, and land cover class as input variables of them, respectively. Results show that over 48% of pixels SPI, and over 32% of pixels SPEI show significantly better consistency with SSMI (at p-value: 0.05). Overall, in regions with cooler temperatures and low and high vegetation cover densities, the SPI, and over warmer areas with mid-range of vegetation density, the SPEI provides a better correlation with SSMI. The performance of different classification methods over validation pixels showed that the K-Nearest Neighbor method can identify the best correlated meteorological drought index with SSMI better than other methods. Overall results highlight the impact of climate and land use on interactions of meteorological and soil moisture drought indices, particularly that the classification efforts showed that, on average, monitoring of drought events by combine use of SPI and SPEI (each one over the area where it is identified as the best indicator of SSMI) improves the correlation between meteorological and soil moisture drought indices by 4 and 10%, compared to the uniform use of SPI and SPEI, respectively.

The Potential of Hyperspectral Image Classification for Oil Spill Mapping

Oil spill mapping is a very challenging problem in marine environmental monitoring. In this paper, the potential of hyperspectral image classification for mapping oil spills is comprehensively investigated. First, several representative hyperspectral image classification methods are reviewed in a general framework. Second, three oil spill mapping cases are designed to analyze the performance of different classification methods in detecting the spatial distribution, classifying the type, and estimating the thickness of oil spills. Finally, the experimental results are analyzed in detail, and some conclusions are given, which bring a comprehensive understanding to scholars who are interested in the fields of hyperspectral remote sensing and oil spill mapping.

Classification of Admission Data Using Classification Learner Toolbox

The analysis of data useful for further process and find the usefulness for the futher. In education sector, Admission is hottest topic. For the counselling of the admission in reputed institution or foreign university, the past data of the person are taken into the consideration. In this Paper, Admission data taken from the reputed Website (Dataset Weblink) and conduct the study of classification methods like Naïve bayes and its accuracy using Matlab. The performance of different classification methods from MATLAB classifier learner toolbox for admission data. Finally, the paper choosen the suitable classification method for the admission data set.

Document image classification: Progress over two decades

Document image classification plays a vital role in the document image processing system. Thus it is of great importance to have a clear understanding of the state-of-the-art of the document image classification field, especially in this deep learning era, which will facilitate the development of effective document image processing systems. In this paper, we provide a comprehensive survey of the progress that has been made in the field of document image classification over the past two decades. We categorize the document images into non-mobile images and mobile images according to the way they are acquired. The existing document image classification methods for these two types of images are reviewed, which are classified as textual-based methods, structural-based methods, visual-based methods and hybrid methods. We further compare the performance of different classification methods on several public benchmark datasets. Finally, we highlight some open issues and recommend promising directions for future research.

Evaluating the performance of different classification methods on medical X-ray images

X-ray images are broadly used for diagnosis and analysis in medical science. In a number of medical applications, such as disease detection, automatic classification systems are very beneficial. Therefore, finding an appropriate classification technique is helpful. In this paper, it is aimed to compare the performance of different classification methods on categorizing X-ray images. In this regard, we have done our experiments in two phases. In phase 1, the effect of wavelet transformation (WT) as a feature selection method is evaluated. Obtaining almost similar performance (about 78%) for probabilistic neural network (PNN) and WT-based PNN, while reducing computational cost, leads us to apply WT on datasets in phase 2. During phase 2, random forest (RF), decision tree (DT), support vector machine (SVM) and Naive Bayes (NB) are applied on obtained images from WT. The results reveal better performance of RF compared to other methods by 82% accuracy. With very close accuracy, SVM got the next place (81%). DT and NB classifiers are the next ones by about 66% accuracy.

Predicting Students’ Academic Procrastination in Blended Learning Course Using Homework Submission Data

Academic procrastination has been reported affecting students' performance in computer-supported learning environments. Studies have shown that students who demonstrate higher procrastination tendencies achieve less than the students with lower procrastination tendencies. It is important for a teacher to be aware of the students' behaviors especially their procrastination trends. EDM techniques can be used to analyze data collected through computer-supported learning environments and to predict students' behaviors. In this paper, we present an algorithm called students' academic performance enhancement through homework late/non-submission detection (SAPE) for predicting students' academic performance. This algorithm is designed to predict students with learning difficulties through their homework submission behaviors. First, students are labeled as procrastinators or non-procrastinators using k-means clustering algorithm. Then, different classification methods are used to classify students using homework submission feature vectors. We use ten classification methods, i.e., ZeroR, OneR, ID3, J48, random forest, decision stump, JRip, PART, NBTree, and Prism. A detailed analysis is presented regarding performance of different classification methods for different number of classes. The analysis reveals that in general the prediction accuracy of all methods decreases with increase in the number of classes. However, different methods perform best or worst for different number of classes.

A Nonlinear Manifold Detection based Model for Software Defect Prediction

Software defect prediction requires developing a new technique which aims at accurately predict defective modules in software system with minimum time and space complexity as well as lesser computational cost. As such, a new model based on Nonlinear Manifold Detection Techniques has been proposed to eliminate undesirable and irrelevant attributes of high dimensional datasets by dimension reduction with more prediction accuracy and improved software quality. In this paper, a novel step towards achieving the goal by developing a new model based on Nonlinear MDTs and comparing its effectiveness with existing Feature Selection techniques for identifying the most accurate defect prediction method. The performance of different classification methods with both new and existing techniques has been evaluated, compared and also tested statistically by using Friedman test and Post Hoc analysis. The result proved that new proposed model based on Nonlinear MDTs is better performance oriented compared to accuracy level of all other techniques.

Statistical Comparison of Classification Algorithms for Medical Datasets

The objective of this study is to compare the statistical performance of different classification methods on three medical datasets in terms of accuracy, recall, precision and F-score. The comparison is performed by building models using 32 different machine learning-based hybridised techniques and classification algorithms, which yields unbiased, accurate and reproducible results. For this purpose, well-known medical datasets; Wisconsin Breast Cancer dataset, Pima Indians Diabetes dataset and Liver Disorders dataset, are used. Different pre-processing approaches such as missing value imputation, normality test and feature selection are implemented in an open source package R for each dataset. The experimental results indicate an average of 10 runs of 10-fold cross-validation. Promisingly, this comprehensive study analyses the statistical performance of 32 classification techniques on each medical datasets.

Deep learning reveals Alzheimer's disease onset in MCI subjects: Results from an international challenge

BackgroundEarly diagnosis of Alzheimer's disease (AD) and its onset in subjects affected by mild cognitive impairment (MCI) based on structural MRI features is one of the most important open issues in neuroimaging. Accordingly, a scientific challenge has been promoted, on the international Kaggle platform, to assess the performance of different classification methods for prediction of MCI and its conversion to AD. New methodThis work presents a classification strategy based on Random Forest feature selection and Deep Neural Network classification using a mixed cohort including the four classes of classification problem, that is HC, AD, MCI and cMCI, to train the model. Moreover, we compare this approach with a novel classification strategy based on fuzzy logic learned on a mixed cohort including only HC and AD. ExperimentsA training set of 240 subjects and a test set including mixed cohort of 500 real and simulated subjects were used. The data included AD patients, MCI subjects converting to AD (cMCI), MCI subjects and healthy controls (HC). This work ranked third for overall accuracy (38.8%) over 19 participating teams. Comparison with existing method(s)The “International challenge for automated prediction of MCI from MRI data” hosted by the Kaggle platform has been promoted to validate different methodologies with a common set of data and evaluation procedures. ConclusionDNNs reach a classification accuracy significantly higher than other machine learning strategies; on the other hand, fuzzy logic is particularly accurate with cMCI, suggesting a combination of these approaches could lead to interesting future perspectives.

Object-Based Method Outperforms Per-Pixel Method for Land Cover Classification in a Protected Area of the Brazilian Atlantic Rainforest Region

Conventional image classification based on pixels hinders the possibilities to obtain information contained in images, while modern object-based classification methods increase the acquisition of information about the object and the context in which it is inserted in the image. The objective of this study was to investigate the performance of different classification methods for land cover mapping in the vicinity of the Alto Ribeira Tourist State Park, a Brazilian Atlantic rainforest area. Two classification methods were tested, including i) a hybrid per-pixel classification using the image processing software ERDAS Imagine version 9.1 and ii) an object-based classification using the software eCognition version 5. In the first method, six different classes were established, while in the second method, another two classes were established in addition to the six classes in the first method. Accuracy assessment of the classification results presented showed that the object-based classification with a Kappa index value of 0.8687 outperformed the per-pixel classification with a Kappa index value of 0.2224. Application of the user's knowledge during the object-based classification process achieved the desired quality; therefore, the use of inter-relationships between objects, superclasses, subclasses, and neighboring classes were critical to improving the efficiency of land cover classification.

Discrimination of edible oils and fats by combination of multivariate pattern recognition and FT-IR spectroscopy: A comparative study between different modeling methods

By using FT-IR spectroscopy, many researchers from different disciplines enrich the experimental complexity of their research for obtaining more precise information. Moreover chemometrics techniques have boosted the use of IR instruments. In the present study we aimed to emphasize on the power of FT-IR spectroscopy for discrimination between different oil samples (especially fat from vegetable oils). Also our data were used to compare the performance of different classification methods. FT-IR transmittance spectra of oil samples (Corn, Colona, Sunflower, Soya, Olive, and Butter) were measured in the wave-number interval of 450–4000cm−1. Classification analysis was performed utilizing PLS-DA, interval PLS-DA, extended canonical variate analysis (ECVA) and interval ECVA methods. The effect of data preprocessing by extended multiplicative signal correction was investigated. Whilst all employed method could distinguish butter from vegetable oils, iECVA resulted in the best performances for calibration and external test set with 100% sensitivity and specificity.

Evaluation of Different Classification Techniques for WEB Data

The growth of data-mining applications such as classification and clustering has shown the need for machine learning algorithms to be applied to large scale data. In this paper we present the comparison of different classification techniques using Waikato Environment for Knowledge Analysis or in short, WEKA. WEKA is open source software which consists of a collection of machine learning algorithms for data mining tasks. The aim of this paper is to examine the performance of different classification methods for a set of large data. The algorithm which have been tested are J48, SMO, Part, OneR, ZeroR and Navies Bayes Algorithm. The Syskill and webert we page rating data [11] with a total data of 1660 and a dimension of 332 rows and 5columns will be used to test and validate the differences between the classification methods or algorithms.

A Permutation Based Procedure for Classification Assessment

This article proposes a permutation procedure for evaluating the performance of different classification methods. In particular, we focus on two of the most widespread and used classification methodologies: latent class analysis and k-means clustering. The classification performance is assessed by means of a permutation procedure which allows for a direct comparison of the methods, the development of a statistical test, and points out better potential solutions. Our proposal provides an innovative framework for the validation of the data partitioning and offers a guide in the choice of which classification procedure should be used

An evaluation of classification models for question topic categorization

We study the problem of question topic classification using a very large real‐world Community Question Answering (CQA) dataset from Yahoo! Answers. The dataset comprises 3.9 million questions and these questions are organized into more than 1,000 categories in a hierarchy. To the best knowledge, this is the first systematic evaluation of the performance of different classification methods on question topic classification as well as short texts. Specifically, we empirically evaluate the following in classifying questions into CQA categories: (a) the usefulness of n‐gram features and bag‐of‐word features; (b) the performance of three standard classification algorithms (naive Bayes, maximum entropy, and support vector machines); (c) the performance of the state‐of‐the‐art hierarchical classification algorithms; (d) the effect of training data size on performance; and (e) the effectiveness of the different components of CQA data, including subject, content, asker, and the best answer. The experimental results show what aspects are important for question topic classification in terms of both effectiveness and efficiency. We believe that the experimental findings from this study will be useful in real‐world classification problems.

Robust centroid based classification with minimum error rates for high dimension, low sample size data

A new method of statistical classification (discrimination) is proposed. The method is most effective for high dimension, low sample size data. It uses a robust mean difference as the direction vector and locates the classification boundary by minimizing the error rates. Asymptotic results for assessment and comparison to several popular methods are obtained by using a type of asymptotics of finite sample size and infinite dimensions. The value of the proposed approach is demonstrated by simulations. Real data examples are used to illustrate the performance of different classification methods.