Vellore Institute of Technology (VIT) is a private university located in&nbsp;Tamil Nadu, India. Founded in 1984, as Vellore Engineering College, the institution offers 20 undergraduate, 34 postgraduate, four integrated and four research programs. It has campuses in Vellore, Amravati, Bhopal and Chennai.

VIT is one of the top ranked private universities in India according to NIRF, THE and QS Rankings.&nbsp;Govt. of India has recognized&nbsp;VIT, Vellore as an&nbsp;Institution of Eminence. This has allowed VIT to take independent quality initiatives and move up in world ranking.

&nbsp;

&nbsp;

VIT University

Department of Software and Systems Engineering

Screening shows impact on cancer mortality rate by decreasing the number of advanced cancers with poor diagnosis, while cancer treatment works through decreasing the case-fatality rate. The prediction of breast cancer survivability has been a challenging research problem for many researchers. The objective of this research work is to propose a Novel model that can analysis the Breast cancer data and do efficient prediction. The contributions made in this paper are as follows, we collected three different the dataset from UCI Machine Learning repositories. We propose an approach, where a detailed comparison made between feature selection algorithms. Trained the datasets using Decision Tree, Random Forest and Support vector machine (SVM) machine learning algorithms. An attempt made to understand the impact of model selection metric in predicting different classes of Brest cancer. The results indicated that the Random forest is the best predictor wit 0.98 accuracy on the holdout sample, SVM came out to be the second with 0.97 accuracy and the Decision Tree came out with 0.96 to be the worst of the four condition tree with 0.95 accuracy. Finally performed prediction using Neural Network with three hidden layers and measured the efficiency, using Root Mean Square Error (RMSE) along with its variations. © 2018 SERSC Australia.

VIT Unitversity

School of Information Technology and Engineering

Vellore Campus

World University root node

International Journal of Grid and Distributed Computing

A Novel Approach to Perform Analysis and Prediction on Breast Cancer Dataset using R

Boosting is the method used to improve the accuracy of any learning algorithm, which often suffers from over fitting problem, because of inappropriate coefficient associated to the data points. The objective of our research is to train the data, such that the weighing error of linear classifier goes to zero and classify the sentiments accurately. In this paper, Gradient ascent approach is used to minimize the weighing error of sentiment classifier by predicting the proper coefficients to the data points in trained dataset. When compared to previous studies on designing a strong classifier, our research is novel in the following areas: Estimation of Maximum Likelihood for logistic regression using Gradient ascent and making use of weights of metric in understanding the behavior of AdaBoost algorithm in classifying the sentiments. In our finding, the first decision stump has training error of 30.44%. After thousand iterations, we observed a smooth transition, where the classification error tends to go down to 8.22% and actually stays at same value. Finally, concluding that Boosting algorithm outperforms Random Forests with lesser Mean squared Test Errors.

International Journal of Intelligent Engineering and Systems

Impact of Gradient Ascent and Boosting Algorithm in Classification

The offshore/on-site teams’ knowledge transfer effectiveness is one of the key predictors for evaluating the outcome of global software development (GSD) project. This study aims at proposing an assessment framework that can be used to evaluate the GSD teams’ (offshore/on-site) knowledge transfer effectiveness in the following criteria: knowledge, team, technology, and organization factors. For evaluating the teams’ knowledge transfer effectiveness in GSD projects, we have integrated the fuzzy decision making trial and evaluation laboratory (DEMATEL) and fuzzy multi-criteria decision making (FMCDM) approaches. Moreover, the proposed DEMATEL-FMCDM methodology is used to determine priority weights of each evaluation criterion and rate knowledge transfer effectiveness on the GSD project outcome from software service outsourcing perspective into two dimensions: product success, (schedule, cost improvement, etc.) and service quality (personal satisfaction, and successful collaboration). The proposed methodology is used for the prediction of a possible rating mechanism for determining the GSD project outcome through knowledge transfer criterions of GSD teams. It also facilitates the determination of knowledge transfer effectiveness of GSD teams’ towards the success of the GSD project outcome and initiates actions to improve the project outcome in GSD environment. This study demonstrates a survey research method that addresses 35 knowledge transfer factors assessed by 70 software professionals at Inowits software organization in India.

Department of Information Security

School of Computer Science and Engineering

International Journal of Machine Learning and Cybernetics

Integration of fuzzy DEMATEL and FMCDM approach for evaluating knowledge transfer effectiveness with reference to GSD project outcome

Face recognition and facial expression recognition using local patterns is the order of the day. Local directional number pattern (LDNP) is one of the prominent descriptor for face recognition. LDNP assigns a 3 bit code for each pixel in the image. The resultant LDNP labeled image is divided into regions to form histogram based descriptor. The histogram bins of all the regions are concatenated to form the final descriptor. In contrast to LDNP, a dimensionality reduced local directional number pattern (DR-LDNP) is proposed in this paper. The proposed descriptor computes single code for each block. This is done by X-ORing of the LDNP codes obtained in a single block. During the process, restructuring of the patterns is done by slightly modifying the LDNP coding pattern constraints. The resultant DR-LDNP descriptor outperforms the existing methods. The experimentation is carried out on standard databases like FERET, YALE, ORL, Cohn–Kannade and JAFFEE and obtained good recognition rates compared to other methods. © 2016, Springer-Verlag Berlin Heidelberg.

Journal of Ambient Intelligence and Humanized Computing

Dimensionality reduced local directional number pattern for face recognition

The accessibility to large amount of datasets in commercial domains has accentuated the importance of data mining in the last few years. Practitioners as well as researchers rely on them to reflect on the magnitude and effect of data-related problems that require solution in business environments. In recent years, the volume of online data submissions (e-commerce data) on products, services and organizations has increased exponentially. However, the submitted data are highly unstructured and largely dependent on language. Mining and extracting useful information from such data is a colossal task, as analysis of the data should include opinion word identification/extraction, aspect extraction and entity extraction. Of the three, the entity extraction is one of the governing approaches in text analysis and plays a major role in e-commerce, biomedical and automobile industries and supports the categorization of the records based on the entity names, generation of short summary on the entities and grouping of the similar records. The existing approaches in entity extraction are capable of recognizing and extracting single-word named entities. However, the product names are often given as a sequence of words (multiple words or multi-word named entities) and, therefore, cannot be recognized by the existing methods. To resolve this issue, this paper presents a novel approach of NLP-Modified Token-based Frequencies of Left and Right (NLP-MTFLR), which is considered as an effective approach to detect and extract the multi-word named products and dominant multi-word named product from the customer review corpus. Using this NLP-MTFLR approach, from the review corpus the subwords and multi-subwords are identified and mapped them with its multi-word named products to recognize dominant product of that corpus. With this dominant product identification, the proposed method reveals in that corpus that the identified dominant product is highly reviewed by the reviewers compared to other products. This NLP-MTFLR approach is achieved 97% accuracy, 77% precision, 89% recall and 82% F-score. © 2017, King Fahd University of Petroleum &amp; Minerals.

Department of Smart Computing

Arabian Journal for Science and Engineering

NLP-MTFLR: Document-Level Prioritization and Identification of Dominant Multi-word Named Products in Customer Reviews

Machine learning is the worldwide recent research technique for various systems as they are intelligent enough to find the solution for classification and prediction problems. The proposed work is about a hybrid genetic fuzzy algorithm that performs an optimal search as well as classification upon uncertain data. The data which is uncertain is suitable for fuzzy classifiers to predict the disease. The hybrid genetic fuzzy system applied on the attributes selects relevant attributes. The selected attributes are fed into the fuzzy classifier. The fuzzy rules are again generated using genetic algorithms. This algorithm is applied on three of the important and bench marking data sets taken from the UCI machine learning repository. The heart disease, Wisconsin breast cancer and Pima Indian diabetes datasets produce classification accuracy as 89.65%, 99.5% and 88.93% respectively. In this article there is a comparative study on few of the feature selection and feature reduction techniques.

Department of Computer Applications

Fulltext

International Journal of Engineering & Technology

Comparative study on dimensionality reduction for disease diagnosis using fuzzy classifier

In the world of web, recommendation system plays a vital role in predicting the user desirable item from a plethora of items. Since the recommendation system generates recommendations by processing the user ratings, the presence of irrelevant or noisy ratings will affect the accuracy of the recommendation system. Malicious users induce noise into the system in order to popularize or to de-promote a particular item which might result in reduced accuracy of recommendations. In addition to malicious users, genuine users offer noisy ratings to the items unknowingly. Despite several of the existing methods to detect and resolve the malicious noise, only few address the existence of natural or non-malicious noise in the recommendation system. Detecting natural noise is not as precise as dealing with the malicious noise. Since natural noise is caused by genuine user, the system needs to classify each rating from the perspective of user rating. In this paper, a fuzzy linguistic approach-based natural noise detection algorithm is proposed to address the uncertainty in detecting the non-malicious noise. The proposed detection algorithm is highly efficient as indicated by our experiments that evaluate noise-induced real-world data sets. © 2018, Taiwan Fuzzy Systems Association and Springer-Verlag GmbH Germany, part of Springer Nature.

International Journal of Fuzzy Systems

A Fuzzy Linguistic Approach-Based Non-malicious Noise Detection Algorithm for Recommendation System

Grid computing environment consist group of heterogeneous resources which provides different services to the user. Grid computing applications require lot of computation power and will perform longer tasks. Present grid computing system has a shape which is inflexible in terms of setting up an ad hoc component. To overcome this herewe proposed an approach introducing Agents with Gridsim simulation toolkit (Grid computing environment).Computational grid has objectives that are to coordinate, share, and distribute different resources to work on one computational job. Gridsim tool kit has the flexibility and extensibility. Agent is hidden in the computer system that is situated in grid environment. This paper improves performance of 75% users and increase transfer rate in grid information service. © 2006-2018 Asian Research Publishing Network (ARPN).

ARPN Journal of Engineering and Applied Sciences

An adaptive fault tolerance mechanism in Grid computing

Clustering of text documents is an important technique for enhancing automated learning. Matching is the technique used in order to relate or match the various set of related documents. Clustering groups a set of documents which are similar and dissimilar for unsupervised learning where the user has learning materials which are from raw data which requires further classification. Established feature extraction strategies intend to change over the representation of the major dimensional data set into a lower-dimensional informational collection by anticipating process through mathematical changes. The concept of feature clustering is to aggregate the features into clusters with a high level of pair wise semantic relations. Each cluster is dealt as a single new feature, and, hence, feature dimensionality can be radically lessened. HAC, K-Means, TF/IDF-weighted vectors and cosine similarities is used for the various vectors of data and is applied to text in a direct way to optimize the vectors. © 2006-2018 Asian Research Publishing Network (ARPN).

Analysing effective methodologies used for text clustering using weighted algorithms

Image enhancement is a quantifying criterion for sharpening and enhancing image quality, where many techniques are empirical with interactive procedures to obtain précised results. The proposed Intensity Histogram Equalization (IHE) approach conquers the noise defects that has a preprocessor to remove noise and enhances image contrast, providing ways to improve the intensity of the image. The preprocessor has the mask production, enlightenment equalization and color normalization for efficient processing of the images which generates a binary image by labeling pixels, overcomes the non-uniform illumination of image and classifies color capacity, respectively. The distinct and discrete mapping function calculates the histogram values and improves the contrast of the image. The performance of IHE is based on noise removal ratio, reliability rate, false positive error measure, Max-Flow Computational Complexity Measure with NDRA and Variation HOD. As the outcome, the different levels of contrast have been significantly improved when evaluated against with the existing systems.