Sangeeta Bhanja Chaudhuri

Research Experiences

I work in the area of Usability Engineering, Data mining and analysis for Occupational health and safety, Applied Visual Communication, HCI and Design, Computational Cognition, Design Research Methods where extreme user research is involved. The work is a conglomeration of visual design, experimental design, research methodology, human computer interaction, behavioral study with the aid of statistical methods, artificial intelligence and machine learning.   I design web pages and interested in visual design.

Ph.D Topic: Techniques for Comprehensibility Evaluation of OHS Sign to ensure better Comprehensibility.

Communication Media (CM) plays a significant role in Occupational Safety and Health (OSH) awareness and training for employees of organizations. Various studies reported that industrial safety pictograms are poorly guessed and hardly understood. Successful communication media design is based on proper evaluation of the design. The industrial signs, which are less comprehensible found in the user study, have been considered for this study. Different food processing industries have been visited, and a survey has been conducted. A user study has been done using research instruments such as a questionnaire and an eye tracking device. The data collected were preprocessed, and four different techniques have been designed to evaluate OSH signs. Subjective (manual), objective (eye-tracking) studies and deep learning methods have been used to evaluate the comprehension of OSH signs. Triangularization (data and methods both ) has been implemented within the study, and the result is quite optimistic. It has been found that there is a positive correlation between all the methods. The output of the designed framework predicts the sign to be comprehensible or non-comprehensible. The developed automated deep learning technique gives more than 80% accuracy in predicting the output. Two publications related to this work can be found in the publication list [A-1], [A-3]. One is under review, and more publication is yet to come.

M.Tech Topic: Design of a modern clinical expert System using data mining algorithms and application of statistical approach for performance evaluation (Title)

The main goal of this study is to design a clinical Expert System that will simulate the work of a medical practitioner. The model developed seems to predict the presence or absence of diseases (cardiovascular disease/ heart disease, premenstrual syndrome (PMS)). Rose2 and MATLAB R2010 have been used for implementing Rough Set Theory and Neural Network. A comparative study was done to record the models' performances. Both the models seem to classify heart and PMS data with encouraging accuracy. The model using a neural network with the BFGS algorithm gave better results than Rough Set Theory. The publication related to this work can be found in the publication list Publications  [A-5]. 

Other Research work during Job Tenure: 

• Automatic Grading of Premenstrual Syndrome: Simulating the Manual Diagnosis Process. In association with Dr. Subhagata Chattopadhyay, G.M. & Head Pharmacy Business, Nationwide Primary Health care Services Pvt. Ltd.

This is a simulation of the diagnosis methodology of the Doctors for the PMS (Pre-Menstrual Syndrome) cases. Several Statistical Techniques like Information Gain, Chi-square Test etc. have been applied to extract the significant symptoms. Feed-Forward Neural network has been used to classify PMS cases with 70% accuracy. A paper has been communicated and it has been published “Automatic Grading of Premenstrual Syndrome: Simulating the Manual Diagnosis Process”. American Journal of Biomedical Engineering, vol. 6(3), pp. 78-85.

• Automatic Recognition of Handwritten Bengali Broken Characters (BBC): Simulating Human Pattern Matching

An automatic detection of handwritten Bengali Broken Characters (BBC) using a feed forward neural network (FFNN) has been proposed. It simulates the Human Visual System (HVS) the way human eye matches the patterns of the broken characters to a meaningful character and identifies it. Here the challenge is to detect and retrieve handwritten character which has been distorted up to 90%. The database consists of fifty Bangla characters, each with twenty samples. Each character is presented as an image, which has been preprocessed, segmented and the features are then extracted. A new method has been proposed in this paper. It uses FFNN to calculate the mismatch for the recognition of a character, where it is observed that the distorted characters show very low mismatch with the original characters. For example, characters up to 70% distortions are found to be retrieved effectively.