Search

Automatic evaluation of transcripts deals with comparing texts and evaluating them using the best available methods in NLP. The core part of this project is to enable the machine to evaluate descriptive transcripts all by itself. This reduces the human effort correcting huge answers in educational institutions and crowd examination platforms. Finding syntax and semantics out of text has been of great effort since years. As a part of this research, many NLP models like entity recognition, POS tagging, and topic modeling like LSA and LDA have come into existence. We are using a combination of these to get the semantics out of a transcript to eventually evaluate it.

Medical record instances can have missing values which makes them unsuitable for learning process. Data Imputation is normally done to fill one or more missing data attribute values. Imputation helps to perform supervised or un-supervised learning after the dataset is free from missing data. Learning process helps the discovery of hidden, valuable and important information that can provide insightful results. Imputation is a data pre-processing task that requires applying distance function to find missing values. In this paper, a distance function named MANTRA is proposed to impute missing data values. The distance function is also called as the imputation measure since it is designed for imputation of missing values. A working example is demonstrated that shows how imputation is achieved using proposed distance function, MANTRA. It is proved that the nominal value that is filled after imputation is same as the original.

In this work, we design and propose an improved fuzzy membership function to detect anomalies and intrusions. The objective of the present approach is to achieve an optimal transformation matrix which can improve classifier accuracies. The transformation matrix is aimed at mapping the original process onto a new fuzzy space, so that the resultant representation is free from noise data and facilitates to improve the overall accuracy and also individual class accuracies. Experimental results show that accuracies obtained using our approach is better compared to other approaches. In particular U2R and R2L accuracies are recorded to be very much promising. This research shows an approach which addresses the improvement in overall accuracy and also improvement in detecting R2L and U2R attack accuracies. In future, we plan to extend this work by applying new measures for dimensionality reduction and classification for improving U2R and R2L attack classification accuracies.