Santanu Ghorai

Nonparallel plane proximal classifier

We observed that the two costly optimization problems of twin support vector machine (TWSVM) classifier can be avoided by introducing a technique as used in proximal support vector machine (PSVM) classifier. With this modus operandi we formulate a much simpler nonparallel plane proximal classifier (NPPC) for speeding up the training of it by reducing significant computational burden over TWSVM. The formulation of NPPC for binary data classification is based on two identical mean square error (MSE) optimization problems which lead to solving two small systems of linear equations in input space. Thus it eliminates the need of any specialized software for solving the quadratic programming problems (QPPs). The formulation is also extended for nonlinear kernel classifier. Our computations show that a MATLAB implementation of NPPC can be trained with a data set of 3 million points with 10 attributes in less than 3s. Computational results on synthetic as well as on several bench mark data sets indicate the advantages of the proposed classifier in both computational time and test accuracy. The experimental results also indicate that performances of classifiers obtained by MSE approach are sufficient in many cases than the classifiers obtained by standard SVM approach.

Automatic Defect Detection on Hot-Rolled Flat Steel Products

Automatic defect detection on hot-rolled steel surface is challenging owing to its localization on a large surface, variation in appearance, and their rare occurrences. It is difficult to detect these defects either by physics-based models or by small-sample statistics using a single threshold. As a result, this problem is focused to derive a set of good-quality defect descriptors from the surface images. These descriptors should discriminate the various surface defects when fed to suitable machine learning algorithms. This research work has evaluated the performance of a number of different wavelet feature sets, namely, Haar, Daubechies 2 (DB2), Daubechies 4 (DB4), biorthogonal spline, and multiwavelet in different decomposition levels derived from 32 × 32 contiguous (nonoverlapping) pixel blocks of steel surface images. We have developed an automated visual inspection system for an integrated steel plant to capture surface images in real time. It localizes defects employing kernel classifiers, such as support vector machine and recently proposed vector-valued regularized kernel function approximation. Test results on 1000 defect-free and 432 defective images comprising of 24 types of defect classes reveal that three-level Haar feature set is more promising to address this problem than the other wavelet feature sets as well as texture-based segmentation or thresholding technique of defect detection.

Cancer Classification from Gene Expression Data by NPPC Ensemble

The most important application of microarray in gene expression analysis is to classify the unknown tissue samples according to their gene expression levels with the help of known sample expression levels. In this paper, we present a nonparallel plane proximal classifier (NPPC) ensemble that ensures high classification accuracy of test samples in a computer-aided diagnosis (CAD) framework than that of a single NPPC model. For each data set only, a few genes are selected by using a mutual information criterion. Then a genetic algorithm-based simultaneous feature and model selection scheme is used to train a number of NPPC expert models in multiple subspaces by maximizing cross-validation accuracy. The members of the ensemble are selected by the performance of the trained models on a validation set. Besides the usual majority voting method, we have introduced minimum average proximity-based decision combiner for NPPC ensemble. The effectiveness of the NPPC ensemble and the proposed new approach of combining decisions for cancer diagnosis are studied and compared with support vector machine (SVM) classifier in a similar framework. Experimental results on cancer data sets show that the NPPC ensemble offers comparable testing accuracy to that of SVM ensemble with reduced training time on average.

Newton's method for nonparallel plane proximal classifier with unity norm hyperplanes

In our previous research we observed that the nonparallel plane proximal classifier (NPPC) obtained by minimizing two related regularized quadratic optimization problems performs equally with that of other support vector machine classifiers but with a very lower computational cost. NPPC classifies binary patterns by the proximity of it to one of the two nonparallel hyperplanes. Thus to calculate the distance of a pattern from any hyperplane we need the Euclidean norm of the normal vector of the hyperplane. Alternatively, this should be equal to unity. But in the formulation of NPPC these equality constraints were not considered. Without these constraints the solutions of the objective functions do not guarantee to satisfy the constraints. In this work we have reformulated NPPC by considering those equality constraints and solved it by Newtons method and the solution is updated by solving a system of linear equations by conjugate gradient method. The performance of the reformulated NPPC is verified experimentally on several bench mark and synthetic data sets for both linear and nonlinear classifiers. Apart from the technical improvement of adding those constraints in the NPPC formulation, the results indicate enhanced computational efficiency of nonlinear NPPC on large data sets with the proposed NPPC framework.

Discriminant Analysis for Fast Multiclass Data Classification Through Regularized Kernel Function Approximation

In this brief we have proposed the multiclass data classification by computationally inexpensive discriminant analysis through vector-valued regularized kernel function approximation (VVRKFA). VVRKFA being an extension of fast regularized kernel function approximation (FRKFA), provides the vector-valued response at single step. The VVRKFA finds a linear operator and a bias vector by using a reduced kernel that maps a pattern from feature space into the low dimensional label space. The classification of patterns is carried out in this low dimensional label subspace. A test pattern is classified depending on its proximity to class centroids. The effectiveness of the proposed method is experimentally verified and compared with multiclass support vector machine (SVM) on several benchmark data sets as well as on gene microarray data for multi-category cancer classification. The results indicate the significant improvement in both training and testing time compared to that of multiclass SVM with comparable testing accuracy principally in large data sets. Experiments in this brief also serve as comparison of performance of VVRKFA with stratified random sampling and sub-sampling.

Multicategory cancer classification from gene expression data by multiclass NPPC ensemble

The discovery of DNA microarray technologies have given immense opportunity to make gene expression profiles for different cancer types. Besides binary classification such as normal versus tumor samples the discrimination of multiple tumor types is also important. In this work, we have first extended the recently developed binary nonparallel plane proximal classifier (NPPC) to multiclass NPPC by decomposition techniques. The multiclass NPPC is then used in a computer aided diagnosis framework to classify multicategory cancer from gene expression data by selecting very few genes by using mutual information criterion. The idea of binary NPPC ensemble is extended to form multiclass NPPC ensemble. Besides usual majority voting method, we have introduced minimum average proximity based decision combiner for multiclass NPPC ensemble. The effectiveness of the proposed method are demonstrated on four benchmark microarray data sets and compared with support vector machine (SVM) classifier in a similar framework.

A Novel Technique of Black Tea Quality Prediction Using Electronic Tongue Signals

Electronic tongue (ET) system is under extensive development for automatic analysis and prediction of quality of different industrial end products. Each sensor in an ET system generates a specific electronic response in presence of different organic or inorganic compounds in the sample. The vital part of the ET system is the discrimination of the complex pattern generated by the sensor array. In this paper, a novel technique of black tea quality estimation is using the ET signals. A moving window is used to extract discrete wavelet transform coefficients from the transient response of ET. The energy in different frequency bands are used as the features of the ET signal for different positions of the window. The prediction of a new sample is performed by the highest score obtained by a particular class by testing all the patterns generated by windowing ET signal. The performance of the proposed technique is verified to estimate black tea quality using two kernel classifiers, namely support vector machine and recently proposed vector valued regularized kernel function approximation method. High prediction accuracy of both the classifiers confirms the effectiveness of the proposed technique of tea quality estimation using ET signals.

Unity norm twin support vector machine classifier

In this work we have reformulated the twin support vector machine (TWSVM) classifier by considering unity norm of the normal vector of the hyperplanes as the constraints. TWSVM with unity norm hyperplanes removes the shortcomings of the classical TWSVM formulation. The resulting new formulation is a nonlinear programming problem which is solved by sequential quadratic optimization method. The performance of the modified classifier verified experimentally on synthetic as well as on benchmark data sets.

Feature Fusion for Prediction of Theaflavin and Thearubigin in Tea Using Electronic Tongue

Liquor characteristics of black cut, tear, and curl tea mostly depend on two biochemical components like theaflavin (TF) and thearubigin (TR). Evaluation of tea quality can be done efficiently by estimating the concentration of TF and TR without using biochemical tests as it takes much time, which requires laborious effort for sample preparation, storage, and measurement. Moreover, the required instruments for this test are very costly. In this paper, we have proposed an efficient method of TF and TR prediction in a given tea sample using electronic tongue (ET) signal. Combinations of transformed features, like discrete cosine transform, Stockwell transform (ST), and singular value decomposition, of ET signals are fused to develop regression models to predict the contents of TF, TR, and TR/TF. Three different regression models such as artificial neural network, vector-valued regularized kernel function approximation, and support vector regression are used to evaluate the performance of the proposed method. High prediction accuracy using fusion of features ensures the effectiveness of the proposed method for prediction of TF and TR using ET signals.

Epithelial mesenchymal transition in lung cancer cells: A quantitative analysis.

Cellular auto-fluorescence along with morphological and cytoskeletal features were assessed in lung cancer cells undergoing induced epithelial mesenchymal transition (EMT). During EMT progression, significant increase was observed in cellular aspect ratio (AR), filamentous (F)-actin and green auto-fluorescence intensities while blue intensity decreased. These features were provided to a kernel classification framework. The classification accuracy were impressive, thus these features along with the classification technique can be considered as suitable tools for automated grading of lung cancer cells undergoing EMT progression.