Harman S.

American hospitals spent over $41 billion on diabetic patients who got readmitted within 30 days of discharge. Being able to determine factors that lead to higher readmission in such patients, and correspondingly being able to predict which patients will get readmitted can help hospitals save millions of dollars while improving quality of care.

We used data mining techniques such as feature engineering, data balancing and interactions, along with robust machine learning models to… Show more

American hospitals spent over $41 billion on diabetic patients who got readmitted within 30 days of discharge. Being able to determine factors that lead to higher readmission in such patients, and correspondingly being able to predict which patients will get readmitted can help hospitals save millions of dollars while improving quality of care.

We used data mining techniques such as feature engineering, data balancing and interactions, along with robust machine learning models to predict medication change and hospital readmission in diabetic patients, and interpret these models to devise corrective measures for hospitals.
Show less

Do people tweet the same way as they vote? With this primary goal in mind, we performed sentiment analysis using Natural Language Processing and Machine Learning on a set of 50,000 tweets to identify the correlation between the tweets and voting behaviors of the voters.

This involved coming up with a subject-based tweet sentiment analyzer that predicted the U.S. presidential election results by state, with an accuracy of 89%. Given any tweet as an input, the web-interface for the… Show more

Do people tweet the same way as they vote? With this primary goal in mind, we performed sentiment analysis using Natural Language Processing and Machine Learning on a set of 50,000 tweets to identify the correlation between the tweets and voting behaviors of the voters.

This involved coming up with a subject-based tweet sentiment analyzer that predicted the U.S. presidential election results by state, with an accuracy of 89%. Given any tweet as an input, the web-interface for the analyzer accurately predicted whether the Tweet was Pro-Hillary or Pro-Trump.

As an extension to the analyzer, we also came up with a visualization of what each of the states within the U.S. was talking about, based on the tweets. Show less

MOOCs have become one of the most popular means of online education in the past few years. However, the biggest problem being faced by MOOC vendors is the significantly high rate of student dropouts.

Having an understanding of which students are going to drop out allows for interventions to be implemented. Moreover, of we can interpret the results of MOOC dropouts, we can improve the ability to tailor the interventions to the respective students who will be most helped by… Show more

MOOCs have become one of the most popular means of online education in the past few years. However, the biggest problem being faced by MOOC vendors is the significantly high rate of student dropouts.

Having an understanding of which students are going to drop out allows for interventions to be implemented. Moreover, of we can interpret the results of MOOC dropouts, we can improve the ability to tailor the interventions to the respective students who will be most helped by them.

Motivated by this, the project involved the use of machine learning techniques such as Recurrent Neural Networks and Ensemble Methods to predict a student's likelihood to drop out of a MOOC on a weekly basis, based on his activities and interaction within the MOOC. Show less

Developed an application to measure packet loss episodes accurately with end-to-end probes in a network. This involved testing the capability of standard Poisson- modulated end-to-end measurements and presenting a more efficient method of packet loss measurement using a self-designed algorithm, that improved the measurement accuracy by approximately 20%.

Developed an application in C that provides a transparent and convenient way of performing the basic file operations, irrespective of the operating system used.

Designed the algorithm for an Image Steganography application (developed in C#) that provides data security by encrypting data files into bitmap image files and subsequently decrypting them, using the encryption-decryption algorithm.

Developed a platform independent application that simulates the DOS environment, while ensuring minimum system requirements.