Qualification: 
Ph.D, M.Tech, BE
pn_kumar@cb.amrita.edu

Dr. (Col.) P. N. Kumar currently serves as Chairperson and Professor at the department of Computer Science and Engineering, Amrita School of Engineering and Head - Students Affairs, Coimbatore Campus. His areas of research include Data Analytics, Business intelligence and Agent based modeling of Financial Markets.

Publications

Publication Type: Book Chapter

Year of Publication Title

2019

Dr. (Col.) Kumar P. N., Parambalath, G., Mahesh, E., and Balasubramanian, P., “Big Data Analytics: A Trading Strategy of NSE Stocks Using Bollinger Bands Analysis”, in Advances in Intelligent Systems and Computin, vol. 839, 2019, pp. 143-154.

2019

M. Srenithi and Dr. (Col.) Kumar P. N., “Motion Detection Algorithm for Surveillance Videos”, in Lecture Notes in Computational Vision and Biomechanics, Book Chapter-2019, vol. 30, Springer Netherlands, 2019, pp. 955-964.[Abstract]


Locality Sensitive Hashing (LSH) is an approach which is extensively used for comparing document similarity. In our work, this technique is incorporated in a video environment for finding dissimilarity between the frames in the video so as to detect motion. This has been implemented for a single point camera archiving, wherein the images are converted into pixel file using a rasterization procedure. Pixels are then tokenized and hashed using minhashing procedure which employs a randomized algorithm to quickly estimate the Jaccard similarity. LSH finds the dissimilarity among the frames in the video by breaking the minhashes into a series of band comprising of rows. The proposed procedure is implemented on multiple datasets, and from the experimental analysis, we infer that it is capable of isolating the motions in a video file.

More »»

2018

K. Sowmya and Dr. (Col.) Kumar P. N., “Traffic Density Analysis Employing Locality Sensitive Hashing on GPS Data and Image Processing Techniques”, in Lecture Notes in Computational Vision and Biomechanics, vol. 28, Springer Netherlands, 2018, pp. 959-971.[Abstract]


Recent development of GPS enabled devices helps in tracking the approximate location of any device. Any GPS enabled device with working internet can be tracked at any point of time. The data obtained from GPS serves several purposes, such as tracking lost devices, providing directions to a certain destination, etc. In several public environments, difficulty arises in plugging the rescue operation during any emergency needs. In case of traffic, the raw data about the traffic closure will not help the authority to reach right location. Instead, the information such as, near accurate location and time of traffic collected from GPS helps the authority to reach the destination. In this paper location of vehicle crowd formation is detected by applying similarity detection with locality sensitive hashing to the collected GPS data and two approaches with LSH (on numerical computation and on image processing) are proposed. The LSH technique is used to hash the location data to find the vehicles at similar locations and time. The proposed approaches are lightweight and needs less computational effort, since dual hashing is employed thus making it suitable for real time applications. This paper uses Apache spark for detecting vehicle crowd by applying LSH to the GPS data, since it is very fast and handles enormous data.

More »»

Publication Type: Journal Article

Year of Publication Title

2019

V. Vismayaa, Pooja, K. R., Alekhya, A., Malavika, C. N., Dr. Binoy B. Nair, and Dr. (Col.) Kumar P. N., “Classifier Based Stock Trading Recommender Systems for Indian stocks: An Empirical Evaluation”, Computational Economics, 2019.[Abstract]


Recommender systems that can suggest the user when to buy and sell stocks can be of immense help to those who wish to trade in stocks but are constrained by their limited knowledge of stock market dynamics. Traditionally, the trading recommendations have been generated on the basis of technical analysis. However, recent research in the field indicates that soft computing/data mining based recommender systems are also capable of generating profitable trading recommendations. An attempt has been made in this study to generate novel classifier based stock trading recommender systems that employ historical stock price data and technical indicators as input features. Moreover, there have been very few studies on the effectiveness recommender systems in the context of India, the world's sixth largest economy and home to one of the world's largest stock exchanges: the Bombay Stock Exchange (BSE). This study presents an empirical evaluation the effectiveness of five single classifier and six ensemble classifier based recommender systems on a total of 293 stocks drawn from the BSE. Recommender system performance for each stock is evaluated based on classification accuracy and eight economic performance measures. Results indicate that the proposed approach can indeed be used successfully for generating profitable trading recommendations.

More »»

2017

T. Rajasundari, Subathra P., and Dr. (Col.) Kumar P. N., “Performance Analysis of Topic Modeling Algorithms for News Articles”, Journal of Advanced Research in Dynamical and Control Systems, vol. 2017, pp. 175-183, 2017.[Abstract]


Topic Modeling is a statistical model, which derives the latent theme from large collection of text. In this work we developed a topic model for BBC news corpus to find the screened regional from the corpus. We have implemented the topic modeling algorithms Latent Dirichlet Allocation (LDA), Latent Semantic Analysis (LSA) and three different machine learning approaches (Naive Bayes, K-NN and K-means). We compared the performance of topic modeling algorithms with machine learning approaches using the measures precision and recall. Our results show that topic modeling algorithms work better for corpus with multiple topic distribution.

More »»

2017

Subathra P., Thiyaneswaran, V., and Dr. (Col.) Kumar P. N., “Preventive System Against Cyber Bulling Using Topic Modeling Algorithm”, Journal of Advanced Research in Dynamical and Control Systems, pp. 287-296, 2017.[Abstract]


Text Mining is one of the techniques used for deriving high quality of information from the text. Topic Modeling is a major task involved in text mining for finding hidden subjects from the corpus. Cyber bullying is the act of bullying people through electronic means of communication which primarily threaten the teens in a deliberate manner. In the proposed work we have taken the survey from students, blogs and YouTube comments which are fed as input to LSA and LDA for finding the cyber bullying documents. Both topic modeling techniques are designed for categorizing the cyber bullying documents from the corpus. We experimented with LSA model for document summarization and using LDA model, developed add-ons on Google Chrome to prevent the cyber bullying words. The Performance analysis is done for both LSA and LDA algorithm. Our results show that LDA performs better for documents with multiple topic distribution.

More »»

2016

A. N., Vijay Krishna Menon, and Dr. (Col.) Kumar P. N., “Cluster Computing Paradigms – A Comparative study of Evolving Frameworks”, IJCTA, (International Conference Soft Computing Systems, ICSCS-2016), vol. 8, pp. 1911-1916, 2016.[Abstract]


Cluster computing is an approach for storing and processing huge amount of data that is being generated. Hadoop and Spark are the two cluster computing platforms which are prominent today. Hadoop incorporates the MapReduce concept and is scalable as well as fault-tolerant. But the limitations of Hadoop paved way for another cluster computing framework named Spark. It is faster and can also manage multiple workloads due to its inmemory processing. In this paper, we discuss the underlying concepts of Hadoop and mention the limitations that led to the development of Spark. Further we give a detailed description about Spark framework and its advantages. We demonstrate a wordcount problem in both Hadoop and Spark and do a comparative study.

More »»

2016

M. Anagha R., Aiswarya, V., Subathra, P., and Dr. (Col.) Kumar P. N., “Browsing Behavioral Analysis Using Topic Modeling”, International Conference Soft Computing Systems”, International Conference Soft Computing Systems”, ICSCS-2016, 2016.

2015

S. Abijith, Renu, P., P. Sachin, S., and Dr. (Col.) Kumar P. N., “A Postulatory Study on Portfolio Optimization Algorithms: A Survey”, Research Journal of Applied Sciences, Engineering and Technology, (RJASET), vol. 11, pp. 988-993, 2015.[Abstract]


A stock represents the capital a company or corporation raises by issuing and subscribing shares. The stock market is a term used to describe the physical location where the buying and selling as well as overall market activity takes place. Companies issue stocks to acquire capital while investors buy them to own a portion of the company. Investors buy stocks with the belief that the company will grow continuously to raise the value of their shares. Every shareholder in a company will have a say on how the company runs. Making investment among various financial enterprises, industries and other categories is associated by a risk factor. Diversification is a technique that is used to mitigate the effects of such risks and creating a portfolio of stocks is the technique used in diversification. In this study, effort has been taken to describe three of the most important portfolio optimization algorithms viz. Genetic Algorithm, Particle Swarm Optimization, Simulated Annealing.

More »»

2015

L. K. Devi, Amrita, R., Subathra P., and Dr. (Col.) Kumar P. N., “An Analysis on the Performance Evaluation of Collaborative Filtering Algorithms Using Apache Mahout”, International Journal of Applied Engineering Research (IJAER), vol. 10, pp. 14797-14812, 2015.[Abstract]


Recommendation systems are being widely adopted in many areas which include social networking, e-commerce etc. Long years of research have led to the proposal of many algorithms in order to effectively capture the real tastes of users and deliver the recommendations accurately. Collaborative filtering is considered to be one of the popular and successful approaches to provide recommendations. In this paper, we conduct a performance evaluation of three popular collaborative filtering algorithms viz. User based, Item based and Slope-one recommender. We illustrate a brief overview on the different approaches of collaborative filtering, their method of working, advantages and limitations. We demonstrate the results based on the evaluation metrics precision, recall, f-measure, fallout and reach. Our experiments revealed that the Slope-one approach outperformed the other two approaches based on the evaluation metrics. We also explored different kinds of similarity metrics and highlighted the effect of size of the neighbourhood on the evaluation metrics. Keywords: Collaborative Filtering (CF), Recommendation systems, Apache Mahout, User based CF, Item based CF, Slope one.

More »»

2015

U. Jaswanth and Dr. (Col.) Kumar P. N., “Big Data Analytics: A Supervised Approach for Sentiment Classification Using Mahout-An Illustration ”, International Journal of Applied Engineering Research (IJAER), vol. 10, pp. 13447-13457, 2015.

2015

S. Jaysri, Priyadharshini, J., Subathra P., and Dr. (Col.) Kumar P. N., “Analysis and Performance of Collaborative Filtering and Classification Algorithms”, International Journal of Applied Engineering Research, vol. 10, pp. 24529-24540, 2015.[Abstract]


Machine learning is a method which is used to learn from data without any human involvement. Recommendation systems come under Machine learning technique which has become one of the essential systems in our day to day e-commerce internet interaction. Many algorithms are proposed to effectively capture the taste of the users and to provide recommendations accurately. Collaborative filtering is one such successful method to provide recommendation to the users. Classification which also falls under Machine learning technique contains many algorithms which can classify text, numerical data, etc. In this paper, we demonstrate two Collaborative Filtering algorithms viz, User based and Item based recommender systems; and three Classification algorithms viz, Naive-Bayes, Logistic Regression and Random Forest Classification. We analysed the results based on evaluation metrics. Our experiment suggests that in Recommender systems, Item based scores over User based; and in Classification, Naive-Bayes emerges superior. © Research India Publications.

More »»

2015

Dr. (Col.) Kumar P. N. and P., B., “Agent based model for market making in emerging stock markets”, International Journal of Applied Engineering Research, vol. 10, pp. 8349-8364, 2015.[Abstract]


Emerging stock markets across the globe are seen to be volatile and also face liquidity problems, vis-à-vis the more matured markets of the developed world. In order to make markets more efficient and an attractive venue for international investors, regulators are required to adopt a suitable market making design. For the purpose of the study on emerging stock markets, the Indian stock market data is analyzed in this paper. Various studies and analyses on the Indian stock markets show that being an emerging market, the volatility of the market is relatively high when compared to that of the other matured developed stock markets like NASDAQ and NYSE. Higher the volatility of a market, higher is the risk involved for investors. Unlike the stock markets of NASDAQ and NYSE, absence of an electronic market maker in the Indian markets appears to be an obvious reason for the prevailing high volatility and this issue is investigated in this paper. The electronic market makers prevalent in the developed markets are generally seen to have a stabilizing effect on the market, apparently reducing volatility to a great extent. This paper demonstrates the suitability of Extended Glosten and Milgrom (EGM) market maker model as the electronic market maker for the Bombay Stock Exchange (BSE) of India. The market maker in the EGM model sets the bid-ask prices based on the orders placed by the traders. It is shown that in the EGM model, the market maker’s quotes reflect the intrinsic value of the stock and any change in the fundamental value (in case of jumps) causes fluctuations in the quotes that are very quickly resolved, thereby bringing stability in the market. The results of the experiments done on the real data from BSE show that, this model can be used as the market maker in the context of any emerging market. This would induce more stability and reduce the volatility, thereby making the market safe for genuine investors. © Research India Publications.

More »»

2013

Dr. (Col.) Kumar P. N., V.P, M., Abinaya, K., Vennilla, S., and Rupika, R., “Implementing an agent based artificial stock market model in JADE - an illustration”, International Journal of Engineering and Technology, vol. 5, pp. 2636-2648, 2013.[Abstract]


Agent-based approach to economic and financial analysis is a suitable research methodolgy for developing and understanding the complex patterns and phenomena that are observed in economic systems. In agent-based financial market models, prices can be endogenously formed by the system itself as the result of interaction of market participants. By using agents for the study, heterogeneous, boundedly rational, and adaptive behaviour of market participants can be analysed and its impact assessed. The collective behaviour of such groups is determined by the interaction of individual behaviours distributed across the group. This being the scenario prevailing in stock markets, agent based models are suitable for the study. Through this paper, we have attempted to illustrate a detailed implementation of multi agents in an artificial stock market invoking the agent-based methodology on Java Agent Development (JADE) environment, a platform to develop multi-agent systems. The Extended Glosten and Milgrom Model, an agent based artificial stock market model, has been chosen to depict the multi-agent environment model in JADE. More »»

2011

Dr. (Col.) Kumar P. N., Amritha, T., Ram, S. V. Gowtham, S. Krishna, H., Karthika, M. N., and Mohandas, V. P., “A Survey of Agent based Artificial Stock Markets (Continuous Session Models)”, International Research Journal of Finance and Economics, vol. 64, pp. 126-139, 2011.[Abstract]


The goal of agent-based modeling of stock markets is to enrich our understanding of fundamental processes that appear in a market. Artificial stock markets are models of financial markets used to study and understand market dynamics. Agent Based Artificial Stock Markets can be seen as any market model in which prices are formed endogenously as a result of participants' interaction. There are various artificial stock markets in existence that are created using different strategies and customized for specific requirements. Trading sessions may be Call market sessions or Continuous sessions. Call market (Discrete) sessions occur at predefined intervals of time whereas trading happens continuously in Continuous sessions. In this paper, we analyze four Continuous Sessions artificial stock market models namely Continuous Time Asynchronous Model (CTAM), Electronic Market Maker (EMM) Model, Continuous Extended Glosten Milgrom Model (CEGM) and KapSyn. In order to facilitate comparison. prior to studying the Continuous Session models, we discuss a Call market (Discrete) session model, the Santa Fe Artificial Stock Market, being one of the heavily cited, first sophisticated agent-based financial market models for studying stock markets, using a bottom-up approach. We analyze their features, design and their pros and cons based on a few important parameters.

More »»

2010

Dr. (Col.) Kumar P. N., G. Seshadri, R., Hariharan, A., Mohandas, V. P., and Dr. P. Balasubramanian, “A Methodology for Aiding Investment Decision between Assets in Stock Markets Using Artificial Neural Network”, International Journal of Computer Science Issues (IJCSI ), vol. 7, no. 6, 2010.[Abstract]


This paper outlines a methodology for aiding the decision making process for investment between two financial market assets (eg a risky asset versus a risk-free asset), using neural network architecture. A Feed Forward Neural Network (FFNN) and a Radial Basis Function (RBF) Network have been evaluated. The model is employed for arriving at a decision as to where to invest in the next time step, given data from the current time step. The time step could be chosen on daily/weekly/monthly basis, based on the investment requirement. In this study, the FFNN has yielded good results over RBF. Consequently the FFNN developed enable us make a decision on investment in the next time step between a risky asset (eg the BSE Sensex itself or a single share) versus a riskfree asset (eg Securities like Govt Bonds, Public Provident Funds etc).The FFNN is trained with a set of data which helps in under standing the market behaviour. The input parameters or the information set consisting of six items is arrived at by considering important empirical features acting on real markets. These are designed to allow both passive and active, fundamental and technical trading strategies, and combinations of these. Using just six items simplifies the decision making process by extracting potentially useful information from the large quantity of historic data. The prediction made by the FFNN model has been validated from the actual market data. This model can be further extended to choose between any two categories of assets whose historical data is available. More »»

2010

Dr. (Col.) Kumar P. N., G. Seshadri, R., Hariharan, A., Mohandas, V. P., and Dr. P. Balasubramanian, “Financial Market Analysis of Bombay Stock Exchange using an Agent Based Model”, International Journal of Imaging Science and Engineering, vol. 8, 2010.[Abstract]


Returns on stocks have traditionally been modelled by fitting a suitable statistical process to empirical returns. Studies on agent based models of stock market have been carried out by researchers, primarily on US markets. This paper analyzes the empirical features generated using historical data from the Bombay Stock Exchange (BSE), employing the concept of agent based model proposed by LeBaron[2,3,8]. Agent-based approach to stock market considers stock prices as arising from the interaction of a number of individual investors. These investors are modeled as intelligent agents, using differing lengths of past information, each trading with its own rules adapting and evolving over time, and this in turn determines the market prices. It is seen that the model generates some features that are similar to those from actual data of the BSE. More »»

Publication Type: Conference Paper

Year of Publication Title

2019

K. N. R. Chebrolu and Dr. (Col.) Kumar P. N., “Deep Learning based Pedestrian Detection at all Light Conditions”, in Proceedings of the 2019 IEEE International Conference on Communication and Signal Processing, ICCSP 2019, 2019, pp. 838-842.[Abstract]


Multispectral pedestrian detection is becoming increasingly important in the field of computer vision due to its applications in driver assistance, surveillance, and monitoring. In this paper, we propose a brightness aware model for pedestrian detection using deep learning. A novel brightness aware mechanism depicts various illumination conditions, so as to enable prediction of day/ night scenario. Based on the detection of the brightness aware mechanism, a color or thermal model is used to detect pedestrians under day or night conditions respectively. The proposed method trained on FLIR-ADAS Thermal dataset and PASCAL VOC Color dataset, has achieved a mAP of '81.27%', which outperforms the current state of the art.

More »»

2016

L. K. Devi, Subathra P., and Dr. (Col.) Kumar P. N., “Performance Evaluation of Sentiment Classification using Query Strategies in a Pool Based Active Learning Scenario”, in Advances in Intelligent Systems and Computing, 2016, vol. 412, pp. 65-75.[Abstract]


In order to perform Sentiment Classification in scenarios where there is availability of huge amounts of unlabelled data (as in Tweets and other big data applications), human annotators are required to label the data, which is very expensive and time consuming. This aspect is resolved by adopting the Active Learning approach to create labelled data from the available unlabelled data by actively choosing the most appropriate or most informative instances in a greedy manner, and then submitting to human annotator for annotation. Active learning (AL) thus reduces the time, cost and effort to label huge amount of unlabelled data. The AL provides improved performance over passive learning by reducing the amount of data to be used for learning; producing higher quality labelled data; reducing the running time of the classification process; and improving the predictive accuracy. Different Query Strategies have been proposed for choosing the most informative instances out of the unlabelled data. In this work, we have performed a comparative performance evaluation of Sentiment Classification in a Pool based Active Learning scenario adopting the query strategies—Entropy Sampling Query Strategy in Uncertainty Sampling, Kullback-Leibler divergence and Vote Entropy in Query By Committee using the evaluation metrics Accuracy, Weighted Precision, Weighted Recall, Weighted F-measure, Root Mean Square Error, Weighted True Positive Rate and Weighted False Positive Rate. We have also calculated different time measures in an Active Learning process viz. Accumulative Iteration time, Iteration time, Training time, Instances selection time and Test time. The empirical results reveal that Uncertainty Sampling query strategy showed better overall performance than Query By Committee in the Sentiment Classification of movie reviews dataset. © Springer Science+Business Media Singapore 2016.

More »»

2015

N. Binoy, Dr. (Col.) Kumar P. N., R., P. Sidharth, Mudra, S. Lopa, Vijayalakshmi, K., Sai, G. R., and J., R., “Forecasting Short-Term Stock Prices using Sentiment Analysis and Artificial Neural Networks”, in International Conference on Soft Computing in Applied Engineering & Sciences (ICSCASE), 2015.

2008

Dr. (Col.) Kumar P. N. and Dr. Isha T. B., “Inductance calculation of 8/6 switched reluctance motor”, in 2008 Joint International Conference on Power System Technology POWERCON and IEEE Power India Conference, POWERCON 2008, New Delhi, 2008.[Abstract]


This paper describes a method to obtain the magnetic characteristics of an 8/6 switched reluctance machine (SRM). Finite element method (FEM) using ANSYS 10.0 software is used for the purpose. Machine model is created by PRO ENGINEER software for the ANSYS environment. To model the machine using FEM, the geometry and mechanicalparameters of the machine are established. Then the model is developed to obtain the flux linkage, current and rotor position relationship, and inductance is computed. Further, this inductance was verified experimentally and found in agreement with each other. © 2008 IEEE. More »»

Publication Type: Conference Proceedings

Year of Publication Title

2017

Dr. (Col.) Kumar P. N. and , “Shuffle Phase Optimization in Spark”, International Conference on Advances in Computing, Communications and Informatics, ICACCI 2017’. pp. 1028-1034 , 2017.

2017

A. John, Dr. Binoy B. Nair, and Dr. (Col.) Kumar P. N., “Application of Clustering Techniques for Video Summarization – An Empirical Study”, Advances in Intelligent Systems and Computing, vol. 573. pp. 494-506, 2017.[Abstract]


Identification of relevant frames from a video which can then be used as a summary of the video itself, is a challenging task. An attempt has been made in this study to empirically evaluate the effectiveness of data mining techniques in video summarization. Video Summarization systems based on histogram and entropy features extracted from three different color spaces: RGB, HSV and YCBCR and clustered using K-Means, FCM, GM and SOM were empirically evaluated on fifty video datasets from the VSUMM [1] database. Results indicate that clustering based video summarizations techniques can be effectively used for generating video summaries

More »»

2015

A. Ravindran, Dr. (Col.) Kumar P. N., and Subathra P., “Similarity Scores Evaluation in Social Networking Sites”, Proceedings of the International Conference on Soft Computing Systems (ICSCS). Springer, 2015.[Abstract]


In today’s world, social networking sites are becoming increasingly popular. Often we find suggestions for friends, from such social networking sites. These friend suggestions help us identify friends that we may have lost touch with or new friends that we may want to make. At the same time, these friend suggestions may not be that accurate. To recommend a friend, social networking sites collect information about user’s social circle and then build a social network based on this information. This network is then used to recommend to a user, the people he might want to befriend. FoF algorithm is one of the traditional techniques used to recommend friends in a social network. Delta-SimRank is an algorithm used to compute the similarity between objects in a network. This algorithm is also applied on a social network to determine the similarity between users. Here, we evaluate Delta-SimRank and FoF algorithm in terms of the friend suggestion provided by them, when applied on a Facebook dataset. It is observed that Delta-SimRank provides a higher precise similarity score because it considers the entire network around a user.

More »»

2015

L. K. Devi, Subathra P., Dr. (Col.) Kumar P. N., V., D. S. Ravi, and B.K., P., “Tweet Sentiment Classification Using an Ensemble of Machine Learning Supervised Classifiers Employing Statistical Feature Selection Methods”, Advances in Intelligent Systems and Computing, vol. 415. Springer Verlag, pp. 1-13, 2015.[Abstract]


Twitter is considered to be the most powerful tool of information dissemination among the micro-blogging websites. Everyday large user generated contents are being posted in Twitter and determining the sentiment of these contents can be useful to individuals, business companies, government organisations etc. Many Machine Learning approaches are being investigated for years and there is no consensus as to which method is most suitable for any particular application. Recent research has revealed the potential of ensemble learners to provide improved accuracy in sentiment classification. In this work, we conducted a performance comparison of ensemble learners like Bagging and Boosting with the baseline methods like Support Vector Machines, Naive Bayes and Maximum Entropy classifiers. As against the traditional method of using Bag of Words for feature selection, we have incorporated statistical methods of feature selection like Point wise Mutual Information and Chi-square methods, which resulted in improved accuracy. We performed the evaluation using Twitter dataset and the empirical results revealed that ensemble methods provided more accurate results than baseline classifiers.

More »»

2010

Dr. (Col.) Kumar P. N., G. Seshadri, R., Hariharan, A., Mohandas, V. P., and Dr. P. Balasubramanian, “Agent based Modeling of Financial Markets”, IEEE International Conference on Computational Intelligence and Computing Research (Best paper award), vol. 2. IEEE Explore , 2010.