Heritage Institute of Technology

About: Heritage Institute of Technology is a based out in . It is known for research contribution in the topics: Support vector machine & Transconductance. The organization has 581 authors who have published 1045 publications receiving 8345 citations.

Multiclass Kernel Classifiers for Quality Estimation of Black Tea Using Electronic Nose

Abstract: Electronic nose (e-nose) is a machine olfaction system that has shown significant possibilities as an improved alternative of human taster as olfactory perceptions vary from person to person. In contrast, electronic noses also detect smells with their sensors, but in addition describe those using electronic signals. An efficient e-nose system should analyze and recognize these electronic signals accurately. For this it requires a robust pattern classifier that can perform well on unseen data. This research work shows the efficient prediction of black tea quality by means of modern kernel classifiers using the e-nose signatures. As kernel classifiers, this work investigates the potential of state of the art support vector machine (SVM) classifier and very recently developed nonparallel plane proximal classifier (NPPC) and vector-valued regularized kernel function approximation (VVRKFA) technique of multiclass data classification to build taster-specific computational models. Experimental results show that VVRKFA and one-versus-rest (OVR) SVM models offer high accuracies to predict the considerable variation in tea quality.

Comparative Study of the Effect of Different Fitness Functions in PSO Algorithm on Band Selection of Hyperspectral Imagery

Abstract: The innate intricacy of hyperspectral images and the absence of the mark data set make the band selection a challenging task in hyperspectral imaging. Computational multifaceted nature can be decreased by distinguishing suitable bands and simultaneously optimizing the number of bands. The PSO (Particle swarm optimization) based technique is used for this purpose. Fitness function takes a significant role in PSO to make a balance between the optimal solution and the accuracy rate. Different distance metrics like Euclidean, City Block, etc. are used as fitness functions and the aftereffects of a similar investigation on different data sets are reported in the present paper.

Identification of Major Depressive Disorder: Using Significant Features of EEG Signals Obtained by Random Forest and Ant Colony Optimization Methods

Abstract: Electroencephalogram (EEG) is an electrophysiological monitoring method to record the electrical activity of the brain. EEG is most often used to diagnose epilepsy, which causes abnormalities in EEG readings. It is also used to diagnose sleep disorders, depth of anesthesia, coma, encephalopathy, brain death, and depression. Being one of the prevalent psychiatric disorders, depressive episodes of major depressive disorder (MDD) is often misdiagnosed or overlooked. Therefore, identifying MDD at earlier stages of treatment could help to facilitate efficient and specific treatment. In this article, Random Forest (RF) and Ant Colony Optimization (ACO) algorithm are used to reduce the number of features by removing irrelevant and redundant features. The selected features are then fed into k-nearest neighbors (KNN) and SVM classifiers, a mathematical tool for data classification, regression, function estimation, and modeling processes, in order to classify MDD and non-MDD subjects. The proposed method used Wavelet Transformation (WT) to decompose the EEG data into corresponding frequency bands, like delta, theta, alpha, beta and gamma. A total of 119 participants were recruited by the University of Arizona from introductory psychology classes based on survey scores of the Beck Depression Inventory (BDI). The performance of KNN and SVM classifiers is measured first with all the features and then with selected significant features given by RF and ACO. It is possible to discriminate 44 MDD and 75 non-MDD subjects efficiently using 15 of 65 channels and 3 of 5 frequency bands to improve the performance, where the significant features are obtained by the RF method. It is found that the classification accuracy has been improved from70.21% and76.67% using all the features to the corresponding 91.67% and 83.33% with only significant features using KNN and Support Vector Machine (SVM) respectively.

Top-K range-aggregate queries on categorical data

Abstract: Efficient processing of top-K queries is a crucial requirement in many interactive applications that deals with huge amount of data. In particular efficient top-K processing has shown a great impact on performance in domains such as the web, text and data integration, business analytics, distributed aggregation of network logs and sensor data, data mining and so on. In this work we focus on some top-K problems in Computational Geometry. We consider a set of problems defined on colored geometric objects (points/intervals) in R1. We are given a set S of n colored geometric objects in R1. Optionally, each object p has a weight w(p) ≥ 0. The number of colors is m ≤ n. For any color c and a query q, let f (c, q) be an aggregation function defined over c-colored objects intersecting q. For a given k ∈ [1 … n], we define top-K(q) to be the set of k colors with the highest k f(c, q) values amongst the distinct colors of the objects intersecting in q. (If the number of distinct colors in q is less than k, we simply include all such colors in top-K(q)). Our goal is to preprocess S into a data structure so that given a query q and an integer k [1 … n], the colors in top-K(q) can be reported efficiently. Efficient solutions are provided to instances of the above general problem in cases where (i) S is a set of colored points, q is a query interval and f(c, q) is the maximum/minimum weight of points of color c in q and (ii) S is a set of colored intervals, q is a query point and f(c, q) is the count of intervals of color c stabbed by q. We use techniques from Computational Geometry to solve these problems.