Venugopal K R

Aspect term extraction for sentiment analysis in large movie reviews using Gini Index feature selection method and SVM classifier

With the rapid development of the World Wide Web, electronic word-of-mouth interaction has made consumers active participants. Nowadays, a large number of reviews posted by the consumers on the Web provide valuable information to other consumers. Such information is highly essential for decision making and hence popular among the internet users. This information is very valuable not only for prospective consumers to make decisions but also for businesses in predicting the success and sustainability. In this paper, a Gini Index based feature selection method with Support Vector Machine (SVM) classifier is proposed for sentiment classification for large movie review data set. The results show that our Gini Index method has better classification performance in terms of reduced error rate and accuracy.

Classification of Neurodegenerative Disorders Based on Major Risk Factors Employing Machine Learning Techniques

Medical data mining has great potential for exploring the hidden patterns in the data sets of the medical domain. These patterns can be utilized for the classification of various diseases. Data mining technology provides a user-oriented approach to novel and hidden patterns in the data. The present study consisted of records of 746 patients collected from ADRC, ISTAART, USA. Around eight hundred and ninety patients were recruited to ADRC and diagnosed for Alzheimers disease (65%), vascular dementia (38%) and Parkinsons disease (40%), according to the established criteria. In our study we concentrated particularly on the major risk factors which are responsible for Alzheimers disease, vascular dementia and Parkinsons disease. This paper proposes a new model for the classification of Alzheimers disease, vascular disease and Parkinsons disease by considering the most influencing risk factors. The main focus was on the selection of most influencing risk factors for both AD and PD using various attribute evaluation scheme with ranker search method. Different models for the classification of AD, VD and PD using various classification techniques such as Neural Networks (NN) and Machine Learning (ML) methods were also developed. It is observed that increase in the vascular risk factors increases the risk of Alzheimers disease. It was found that some specific genetic factors, diabetes, age and smoking were the strongest risk factors for Alzheimers disease. Similarly, for the classification of Parkinsons disease, the risk factors such as stroke, diabetes, genes and age were the vital factors.

Feature Level Fusion Based Bimodal Biometric Using Transformation Domine Techniques

Bimodal biometric used to authenticate a person is more accurate compared to single biometric trait. In this paper we propose Feature Level Fusion based Bimodal Biometric using Transformation Domine Techniques (FLFBBT). The algorithm uses two physiological traits viz., Fingerprint and Face to identify a person. The Region of Interest (ROI) of fingerprint is obtained using preprocessing. The features of fingerprint are extracted using Dual Tree Complex Wavelet Transforms (DTCWT) by computing absolute values of high and low frequency components. The final features of fingerprint are computed by applying log on concatenated absolute value of high and low frequency components. The face image is preprocessed by cropping only face part and Discrete Wavelet Transforms (DWT) is applied. The approximation band coefficients are considered as features of face. The fingerprint and face image features are concatenated to derive final feature vector of bimodal biometric. The Euclidian Distance (ED) is used in matching section to compare test biometric in the database, it is observed that the values of EER and TSR are better in the case of proposed algorithm compared to individual transformation domain techniques.

MSIGT: Most Significant Index Generation Technique for cloud environment

Cloud Computing is a computing paradigm for delivering computational power, storage and applications as services via Internet on a pay-as-you-go basis to consumers. The data owner outsources local data to the public cloud server to reduce the cost of the data management. Critical data has to be encrypted to ensure privacy before outsourcing. The state-of-the-art SSE schemes search only over encrypted data through keywords, hence they do not provide effective data utilisation for large dataset files in cloud. We propose a Most Significant Index Generation Technique (MSIGT), that supports secure and efficient index generation time using a Most Significant Digit (MSD) radix sort. MSD radix sort is simple and faster in sorting array strings. A mathematical model is developed to encrypt the indexed keywords for secure index generation without the overhead of learning from the attacker/cloud provider. It is seen that the MSIGT scheme can reduce the cost of data on owner side to O(NT × 3) with a score calculation of O(NT). The proposed scheme is effective and efficient in comparison with the existing algorithms.

HSAS: Hindi Subjectivity Analysis System

With the development of Web 2.0, we are abundant with the documents expressing users opinions, attitudes and sentiments in the textual form. This user generated textual content is an important source of information to make sound decisions by the organizations and the government. The textual information can be categorized into two types: facts and opinions. Subjectivity analysis is the automatic extraction of subjective information from the opinions posted by users and divides the content into subjective and objective sentences. Most of the works in subjectivity analysis exists for English language data but with the introduction of unicode standards UTF-8, Hindi language content on the web is growing very rapidly. In this paper, Hindi Subjectivity Analysis System (HSAS) is proposed. It explores two different methods of generating subjectivity lexicon using the available resources in English language and their comparative evaluation in performing the task of subjectivity analysis at the sentence level. The first method uses English language OpinionFinder subjectivity lexicon. The second method uses a small seed word list of Hindi language and expands it to generate subjectivity lexicon. Different evaluation strategies are used to validate the lexicon. We achieved 71.4% agreement with human annotators and ~80% accuracy in classification on a parallel data set in English and Hindi. Extensive simulations conducted on the test dataset confirm the validity of the suggested method.

Mobile cloud computing for medical applications

Mobile Devices like Smartphones and Tablets are getting ever increasing processing power. This makes them powerful enough to do heavy duty realtime 3D video processing. Tablets capable of recording 3D video using stereo camera can have applications in various fields. In the medical field these tablets can be used for micromanipulations such as microsurgery. We describe the design of a tablet capable of 3D vision and a glasses free stereoscopic display, which can be used to enhance the working of the technicians performing micromanipulations. We also illustrate how by providing a deep integration with the cloud, we can reduce the processing overheads and enable valuable new services. For example, in case of microsurgeries, the process is very specialized and is done in a few specialized medical centers only. When these powerful mobile devices are coupled with the cloud, it adds whole new dimension in the services which can be offered. With our deep cloud integration, the surgery being performed can be streamed in realtime from the Tablet itself. This live feed can be viewed by a remote expert to provide guidance to the surgeon. The students at remote locations can also view this live feed. This would provide them a unique learning opportunity. Moreover, the recordings of these surgeries can be backed up on cloud and provided for on demand viewing for learning purposes.

Probabilistic Model for Single and Multi-Sensing Intrusion Detection in Wireless Sensor Networks

Wireless Sensor Networks consists of tiny devices capable of processing, routing the sensed data and are capable of detecting the intruders. The process of detecting any suspected(anomalous) moving object(attacker) within the reach of a Wireless Sensor Network area is referred to as intrusion detection. In this paper, we propose an algorithm to detect the intruder by the cluster heads in a 2D and 3D homogeneous Wireless Sensor Networks. This algorithm overcomes the attacks on implementation and also, reduces the energy consumption. The proposed algorithm considers Single Sensing and Multi-Sensing Intrusion Detection using minimum number of sensor nodes and a probabilistic model has been developed for both 2D and 3D homogeneous networks. Simulation results show that the power analysis attack and energy consumption is minimized by activating only few sensor nodes for detection and using only few sensor nodes for processing of data. The performance of the proposed algorithm is better compared to using all the sensor nodes for detection where the energy consumption is more.

Efficient video transfer using LAN caching assisted by cloud computing

There is a good probability of accessing same video content multiple times from a cloud based Video Streaming Server by same peer or different peers of a given LAN, effectively increasing Internet bandwidth or data flow for same content from server to client, thereby over loading routers between server and client and also resulting in higher power consumption at routers. This proposed concept tries to avoid multiple streaming of high volume video files from Server by caching first successful streamed data on to LAN peer which is currently viewing the video data and subsequently the same LAN peer streaming the video to other desiring peers when demanded for. Proposed implementation model retains all other server activities with server except for allowing an available LAN peer copy of video to be streamed to another peer of the same LAN when requested for.

SDLM: Source detection based local monitoring in wireless sensor networks

Security in wireless sensor networks is critical due to its way of open communication. Local monitoring is one of the powerful technique to secure the data and detect various malicious activities. In local monitoring, neighbour nodes observe the communication between current sender, current receiver and next hop receiver to detect the malicious activity. To make sensors power efficient, sleep-wake scheduling algorithms along with local monitoring are suggested in literature. Solutions in the literature do not address the problem if source node is malicious and do not consider unnecessary wake up of the nodes as malicious activity. This paper tries to achieve without assuming source node as honest and considers unnecessary wake up of the node as a malicious activity. Simulated the algorithm in NS-2 and performance analysis is discussed. Even with additional checks applied to detect malicious activities, analytical results show no degradation in the performance.

Convolution based Face Recognition using DWT and feature vector compression

Face Recognition is important Biometric credentials for identification or verification of a person. In this paper, we propose a novel technique of generating compressed unique features of face images which helps in improving matching speed of recognition. The training face database samples are applied to 2D-DWT to obtain LL band features. The LL band features are subjected to normalization to scale the magnitude values in the range 0 to 1. The output of normalization is further convolved with the original face sample to obtain unique features. The convolved output is subjected to Gaussian filter to obtain smoothened image features. Further, The feature vector of several image samples of single person are compressed to convert into single vector to database feature vectors are created by compressing feature vectors of single person face samples in to single column unique vectors which helps in scaling down of feature vectors and improve matching speed. The test samples are subjected to same process to generate unique compressed test feature vectors and are compared with database vectors using Euclidean distance. The results are tabulated for different set of face databases and also compared with existing techniques to validate the performance of proposed method.