Terms Of Misclassification Error Research Articles

Multivariate classification of cannabis chemovars based on their terpene and cannabinoid profiles

Cannabis is used to treat various medical conditions, and lines are commonly classified according to their total concentrations of Δ9-tetrahydrocannabinol (THC) and cannabidiol (CBD). Based on their ratio of total THC to total CBD, cannabis cultivars are commonly classified into high-THC, high-CBD, and hybrid classes. While cultivars from the same class have similar compositions of major cannabinoids, their levels of other cannabinoids and their terpene compositions may differ substantially. Therefore, a more comprehensive and accurate classification of medicinal cannabis cultivars, based on a large number of cannabinoids and terpenes is needed. For this purpose, three different chemometric-based classification models were constructed using three sets of chemical profiles. We examined those models to determine which provides the most accurate “chemovar” classification. This was done by analyzing profiles of cannabinoids, terpenes, and the combination of these substances using the partial least square-discriminant analysis multivariate (PLS-DA) technique. The chemical profiles were selected from the three major classes of medicinal cannabis that are most commonly prescribed to patients in Israel: high-THC, high-cannabigerol (CBG), and hybrid. We studied the correlations between cannabinoids and terpenes to identify major bio-indicators representing the plant's terpene and cannabinoid content. All three PLS-DA models provided highly accurate classifications, utilizing six to nine latent variables with an overall accuracy ranging from 2 to 11% CV. The PLS-DA model applied to the combined cannabinoid-and-terpene profile did the best job of differentiating between the chemovars in terms of misclassification error, sensitivity, specificity, and accuracy. The combined cannabinoid-and-terpene PLS-DA profile had cross-validation and prediction misclassification errors of 4% and 0%, respectively. This is the first study to demonstrate the highly accurate classification of samples of medicinal cannabis based on their cannabinoid and terpene profiles, as compared to cannabinoid profiles alone. Furthermore, our correlation analysis indicated that 11 cannabinoids and terpenes might serve as bio-indicators for 32 different active compounds. These findings suggest that the use of multivariate statistics could assist in breeding studies and serve as a tool for minimizing the mislabeling of cannabis inflorescences.

Image Analysis of Sauvola and Niblack Thresholding Techniques

Image segmentation is a critical problem in computer vision and other image processing applications. Image segmentation has become quite challenging over the years due to its widespread use in a variety of applications. Image thresholding is a popular image segmentation technique. The segmented image quality is determined by the techniques used to determine the threshold value.A locally adaptive thresholding method based on neighborhood processing is presented in this paper. The performance of locally thresholding methods like Niblack and Sauvola was demonstrated using real-world images, printed text, and handwritten text images. Threshold-based segmentation methods were investigated using misclassification error, MSE and PSNR. Experiments have shown that the Sauvola method outperforms real-world images, printed and handwritten text images in terms of misclassification error, PSNR, and MSE.

Predicting tweet impact using a novel evidential reasoning prediction method

This study presents a novel evidential reasoning (ER) prediction model called MAKER-RIMER to examine how different features embedded in Twitter posts (tweets) can predict the number of retweets achieved during an electoral campaign. The tweets posted by the two most voted candidates during the official campaign for the 2017 Ecuadorian Presidential election were used for this research. For each tweet, five features including type of tweet, emotion, URL, hashtag, and date are identified and coded to predict if tweets are of either high or low impact. The main contributions of the new proposed model include its suitability to analyse tweet datasets based on likelihood analysis of data. The model is interpretable, and the prediction process relies only on the use of available data. The experimental results show that MAKER-RIMER performed better, in terms of misclassification error, when compared against other predictive machine learning approaches. In addition, the model allows observing which features of the candidates’ tweets are linked to high and low impact. Tweets containing allusions to the contender candidate, either with positive or negative connotations, without hashtags, and written towards the end of the campaign, were persistently those with the highest impact. URLs, on the other hand, is the only variable that performs differently for the two candidates in terms of achieving high impact. MAKER-RIMER can provide campaigners of political parties or candidates with a tool to measure how features of tweets are predictors of their impact, which can be useful to tailor Twitter content during electoral campaigns.

Information preserving histogram segmentation of low contrast images using fuzzy measures

This paper proposes an improved histogram segmentation technique. In order to preserve the entropy of images with low contrast, the normalisation function has been performed. Qualitative observations made in present study suggest that the loss in entropy of the images may lead to improper histogram segmentation. As per quantitative investigations, based on the simulation results, the proposed method outperforms other contemporary methods in terms of misclassification error (ME), sensitivity (S) and specificity (Sp). The proposed method has been effectively trained and tested over a data set of 80 and 130 images respectively. Statistical consistency of the results is verified using ANOVA test.

A Tool for Classification and Regression Using Random Forest Methodology: Applications to Landslide Susceptibility Mapping and Soil Thickness Modeling

Classification and regression problems are a central issue in geosciences. In this paper, we present Classification and Regression Treebagger (ClaReT), a tool for classification and regression based on the random forest (RF) technique. ClaReT is developed in Matlab and has a simple graphic user interface (GUI) that simplifies the model implementation process, allows the standardization of the method, and makes the classification and regression process reproducible. This tool performs automatically the feature selection based on a quantitative criterion and allows testing a large number of explanatory variables. First, it ranks and displays the parameter importance; then, it selects the optimal configuration of explanatory variables; finally, it performs the classification or regression for an entire dataset. It can also provide an evaluation of the results in terms of misclassification error or root mean squared error. We tested the applicability of ClaReT in two case studies. In the first one, we used ClaReT in classification mode to identify the better subset of landslide conditioning variables (LCVs) and to obtain a landslide susceptibility map (LSM) of the Arno river basin (Italy). In the second case study, we used ClaReT in regression mode to produce a soil thickness map of the Terzona catchment, a small sub-basin of the Arno river basin. In both cases, we performed a validation of the results and a comparison with other state-of-the-art techniques. We found that ClaReT produced better results, with a more straightforward and easy application and could be used as a valuable tool to assess the importance of the variables involved in the modeling.

Distance-based variable generation with applications to the FACT experiment

We introduce a new way to construct variables for classification in a setting of astronomy. The newly constructed variables complement the currently used Hillas parameters and are specifically designed to improve the classification. They are based on fitting elliptic or skewed bivariate distributions to images gathered by imaging atmospheric Cherenkov telescopes and evaluating the distance between the observed and the fitted distribution. As distance measures we use the Chi-square distance, the Kullback–Leibler divergence and the Hellinger distance. The new variables lead to an improved classification in terms of misclassification errors.

SECURED DISCLOSURE OF SENSITIVE DATA IN DATA MINING TECHNIQUES

Recent advances in data collection, data dissemination and related technologies have inaugurated a new era of research where existing data mining algorithms should be reconsidered from the point of view of securing sensitive data. People have become increasingly unwilling to share their data. This frequently results in individuals either refusing to share their data or providing incorrect data. In turn, such problems in data collection can affect the success of data mining, which relies on sufficient amounts of accurate data in order to produce meaningful results. Based on the analysis of shortcomings of earlier technologies this study proposes a new method for securing numerical and categorical data. In this method the categorical data is converted into Binary form and perturbation based noise is introduced as a security method based on the security level anticipated. Several types of noise addition methods were employed and generalized results were evaluated in terms of misclassification error and privacy level. An average of misclassification error was below 50% for 75-90% security level, which is better than earlier methods which didn’t handle categorical data. The results obtained prove that the proposed method outperforms some of the currently existing methods thereby ensuring the possibility of securing sensitive data irrespective of its type being numerical or categorical.

Elastic SCAD as a novel penalization method for SVM classification tasks in high-dimensional data

BackgroundClassification and variable selection play an important role in knowledge discovery in high-dimensional data. Although Support Vector Machine (SVM) algorithms are among the most powerful classification and prediction methods with a wide range of scientific applications, the SVM does not include automatic feature selection and therefore a number of feature selection procedures have been developed. Regularisation approaches extend SVM to a feature selection method in a flexible way using penalty functions like LASSO, SCAD and Elastic Net.We propose a novel penalty function for SVM classification tasks, Elastic SCAD, a combination of SCAD and ridge penalties which overcomes the limitations of each penalty alone.Since SVM models are extremely sensitive to the choice of tuning parameters, we adopted an interval search algorithm, which in comparison to a fixed grid search finds rapidly and more precisely a global optimal solution.ResultsFeature selection methods with combined penalties (Elastic Net and Elastic SCAD SVMs) are more robust to a change of the model complexity than methods using single penalties. Our simulation study showed that Elastic SCAD SVM outperformed LASSO (L1) and SCAD SVMs. Moreover, Elastic SCAD SVM provided sparser classifiers in terms of median number of features selected than Elastic Net SVM and often better predicted than Elastic Net in terms of misclassification error.Finally, we applied the penalization methods described above on four publicly available breast cancer data sets. Elastic SCAD SVM was the only method providing robust classifiers in sparse and non-sparse situations.ConclusionsThe proposed Elastic SCAD SVM algorithm provides the advantages of the SCAD penalty and at the same time avoids sparsity limitations for non-sparse data. We were first to demonstrate that the integration of the interval search algorithm and penalized SVM classification techniques provides fast solutions on the optimization of tuning parameters.The penalized SVM classification algorithms as well as fixed grid and interval search for finding appropriate tuning parameters were implemented in our freely available R package 'penalizedSVM'.We conclude that the Elastic SCAD SVM is a flexible and robust tool for classification and feature selection tasks for high-dimensional data such as microarray data sets.

Texture based segmentation of epithelial layer from oral histological images

The objective of this paper is to provide a texture based segmentation algorithm for better delineation of the epithelial layer from histological images in discriminating normal and oral sub-mucous fibrosis (OSF). As per literature and oral clinicians, it is established that the OSF initially originates and propagates in the epithelial layer. So, more accurate segmentation of this layer is extremely important for a clinician to make a diagnostic decision. In doing this, Gabor based texture gradient is computed in gray scale images, followed by preprocessing of the microscopic images of oral histological sections. On the other hand, the color gradients of these images are obtained in the transformed Lab color space. Finally, the watershed segmentation is extended to segment the layer based on the combination of texture and color gradients. The segmented images are compared with the ground truth images provided by the oral experts. The segmentation results depict the superiority of the texture based segmentation in comparison to the Otsu's based segmentation in terms of misclassification error. Results are shown and discussed.

Evaluating predictive quality models derived from software measures: Lessons learned

This paper describes an empirical comparison of several modeling techniques for predicting the quality of software components early in the software life cycle. Using software product measures, we built models that classify components as high-risk, i.e., likely to contain faults, or low-risk, i.e., likely to be free of faults. The modeling techniques evaluated in this study include principal component analysis, discriminant analysis, logistic regression, logical classification models, layered neural networks, and holographic networks. These techniques provide a good coverage of the main problem-solving paradigms: statistical analysis, machine learning, and neural networks. Using the results of independent testing, we determined the absolute worth of the predictive models and compare their performance in terms of misclassification errors, achieved quality, and verification cost. Data came from 27 software systems, developed and tested during three years of project-intensive academic courses. A surprising result is that no model was able to effectively discriminate between components with faults and components without faults.