• Quiz App

The application allows the user to choose a category of their choice – vocabulary, chemistry, history, physics to learn the terms used in that field. The user can choose “Warm Up” to revise the words, “Work Out” to solve the timed quiz or the “Marathon”mode to complete a longer stretch of questions.

Concepts Implemented : SQLite Database, List/ Card View, Firebase, JSON, Testing, Network.

GitHub : Quiz App

• Multi-label Classification on Yelp Dataset

To predict the star ratings for the Yelp businesses using the business reviews, total check-ins, state and city. This project utilizes the 5.5 GB Yelp data and compares different Machine learning models. It also compares the Spark Ml libraries and Python scikit machine learning libraries.

GitHub: Multi-label Classification.

• H1B Classifier

To study of the impact of parameters on different models in ML multiclass classification algorithms on H1B Dataset on Kaggle to find the case status of the visa petition in Spark.

The graph below indicates a comparative study between the accuracy, precision and recall for all 4 models. 

Accuracy	We observe the lowest accuracy in Naive Bayes Model, then come both Logistic Regression and Decision Tree. The highest accuracy is obtained in Random Forest Model.
Precision	The highest precision of 100% is achieved by the Logistic Regression model. Almost near it is Random Forest with a precision of 99.99%. Decision Tree has a precision of 99.02% while Naive Bayes has the lowest precision of 52.48%.
Recall	The highest recall rate achieved is that of Random Forest Model. After which Logistic Regression and Decision Tree come at the same scale. And Naive Bayes has the lowest recall rate.
Runtime	The runtime is highest at 194.742 secs for Random Forest, Decision Tree at 32.03 secs, then is Logistic Regression at 25.55 secs. Best runtime achieved is that by Naive Bayes at 9.913 secs.

• Locator IOS App

The application allows the user to post where they are from and allow them to see on a map where others are from. The application has two basic parts. First part is posting the user information and second is viewing of data posted by other users.

Concepts Implemented : Map and list view, geocoding, RESTful API, Cocoa Pods.

GitHub: Locator.

• Credit Card Fraud Detection  

Problem Statement : As credit card becomes the most popular mode of payment for both online as well as regular purchase, cases of fraud associated with it are also rising. Fraud is one of the major ethical issues in the credit card industry.

Solution : A software system for banks to detect frauds on basis of comparison with the database of the customer which includes location, frequency of transactions, amount and the trend of user expenditure using Genetic algorithm with an online shopping frontend and java applet backend.

Sample result :