This project was to design a navigation system for the visually impaired. Since this has been a problem that has gone through research over the years we had to learn and apply some of the previous works done by researchers to solve this problem
For the first task, we detected doors based on the geometry and heuristics that would help us detect doors within an indoor environment. We used a 3 D camera so that we could accurately capture the depth and distance information. With this we were able to detect whether the doors are open or closed.
The second task was based on the objective to detect obstacles other than doors within an indoor environment. This task was based on K means segmentation ,sift points detection , sift localisation and contour grouping. With this we were able to detect obstacles and obtain distance metrics.
The objective of this project is to predict somatic mutation, given a protein sequence using a feature based classifier. This is a neutral vs somatic mutation binary classification problem. Somatic mutations can be divided into passenger and driver mutations. Driver mutations are considered as cancer-causing mutations and an analysis of features affecting somatic vs neutral classification can be a basis of the task of distinguishing between driver and passenger mutations to predict cancer with the mutation information.
The project was based on the impending danger of exceeding Opiate drug overdose. The project was based on the objective to predict whether a prescriber is Opiate prescriber or not based on their non opiate prescription pattern. The recent drug overdose death due to Opiate drug intake was the main push for us to work on this Project.
This project is a twitter sentimental analysis project based on the 2016 US election data. The main objective for us was to predict the chances of the electoral candidates HillaryClinton and Donaldtrump on various US states
The data extraction process for this project was critical as we had to crawl data using the Twitter API. The data extraction process involved collecting data that tweeted about the twitter mention tags @realDonaldTrump and @HillaryClinton. We were able to extract data from different parts of the world for our project. Since our main objective was to concentrate on the sentiments of the people in US we stuck to the tweets tweeted from different parts of the US. We were then able to create visualizations using cartograms and other visualization methods to intuitively understand the sentiments of the people from different states in the US.
This project works on the hypothesis of a citation recommendation system.
If two papers share a similar title, then the papers being cited would also contain a common theme / topic. It is because of this hypothesis that we calculate the similarity of titles and provide similar titled papers as inputs for collaborative filtering.
We find that titles as Bag of Words representation by itself cannot have enough information to retrieve similar papers. We introduce a concept called "context citations". This concept is similar to word2vec's philosophy "The word is explained by the company that it chooses", we propose " A paper is characterized by the paper that it cites" more than the title similarity. This is also a smart way of avoiding the cold start problem by giving the initial inputs for the recommender system.
This project is an online challenge based on the Yelp dataset.
For the first task, the dataset was downloaded from Yelp and the process of stemming and stop words removal was done on the reviews of the customers. From this, we generated a unigram, bigram and trigram model. A tf -idf weighting was done to calculate the weights of words. With these our sole aim is to predict the restaurant category the review falls under. The Naive Bayes and LSH algorithms were used for this task.
For this task we gather the customer visits to each restaurants from the source files. We then use Matrix factorisation methods with solvers like Alternate Least Squares (ALS) and Stochastic Gradient Descent (SGD). The idea here is to build a recommednation system that would predict on which restaurant the user would visit next.
This project is a three tier telecom system project developed using C#. The project provided me a good head start with C#. I was able to design an effective front end and implement necessary validations and understand the working of C# by developing this project.