Search

There is a large variation in EEG signals from human to human. Therefore, it is a tough task to create a subject-independent emotion recognition system using EEG. EEG is reliable than facial expression or speech signal to recognize emotions, since this cannot be self-produced. The proposed study aims to develop a subject-independent emotion recognition system with a benchmark database DEAP. In this work, deep neural network with simple architecture is used to classify low–high valence and similarly low–high arousal. EEG signals are nonstationary signals. In this, the stochastic properties as well as spectrum changes over time. For these types of signals, the wavelet transform would be suitable as features, hence, wavelet transform is used to obtain different frequency bands of EEG signals.

EEG signals vary from human to human and hence it is very difficult to create a subject independent emotion recognition system. Even though subject dependent methodologies could achieve good emotion recognition accuracy, the subject-independent approaches are still in infancy. EEG is reliable than facial expression or speech signal to recognize emotions, since it can not be fake. In this paper, a Multilayer Perceptron neural network based subject-independent emotion recognition system is proposed. Performance evaluation of the proposed system, on the benchmark DEAP dataset shows good accuracy compared to the state of the art subject independent methods.

Handwriting is one of the behavioral biometric techniques for person identification. Our paper proposes a novel grapheme-based handwriting recognition system. First, the images of handwritings are segmented into graphemes and then graphemes are represented as projection profile. Then, dictionary is learnt from the images of handwriting using k-means clustering. Now using dictionary, writer feature vectors are generated and stored with writer label. Finally, query document features are matched with writer feature vectors using k-nearest neighbor. The outcome is the writer label with the highest similarity score.

As establishing fraud detection mechanism in the healthcare industry is an evolving challenge, this research work proposes a comprehensive approach for predicting the most probable fraudulent claims by the help of traditional and advanced statistical data analysis tools as well as algorithms. Keeping in mind the business value of the fraudulent claims, the research work seeks to develop a systemic methodology of modeling the information in such a way that the defined business goals in healthcare fraud management are achieved. Given the nature of insurance industry, the currently available mechanisms lack a standardized approach for encountering the events of fraudulent activities. So the research area focuses on the use of a distinguished methodology of “Rules-based scoring system” and creating a well-defined statistical model to assess the correlating factors which describe the probability and propensity of fraudulent claims in the industry. The research work leads to develop various data models using algorithms such as logistic regression, neural network, and decision tree in order to understand the cohesion between probable factors and also build up the most accurate model which can address or predict the claims based on the prevailing conditions and transactional activities.