Snapchat Filter Using CNN and OpenCV
The main objective of this project is to overlay the Snapchat-like filters by locating and resizing the filters according to the individual’s predicted the facial keypoints. In this project, I used pre-trained frontal face haarcascade models provided by OpenCV for real-time face detection and publicly available dataset ‘Facial Keypoints Detection’ from Kaggle to predict the keypoints positions for the unseen face images. I used convolutional neural networks since they are powerful for solving prediction problems involving image data as input. The model was trained and was able to reach a minimum loss of approximately 0.002 and an accuracy of 80%. In the final step, the filter was placed slightly above the predicted eyebrow points and the width of the filter was determined using the top left and right eyebrow points. For more explanation of the project, please read this article:
Airbnb and Zillow Analysis
I developed a data product to help real estate clients understand which zip codes would generate the most
profit on short term rentals within New York City to make an investment. Two publicly available datasets
from Airbnb and Zillow were used. Airbnb data consisted of information about properties, host, availability,
rental price in certain neighborhoods in New York City, etc. Zillow provided cost estimate for
two-bedroom properties.
The quality of data was checked, preprocessed and the insights were highlighted followed by data wrangling
and munging steps. Few key metrics were derived in order to evaluate the most profitable
zipcode. FbProphet model was used for forecasting cost of property for the years 2018 to 2022.
The key metrics and visualizations led us to gain insights about the properties which were helpful
in making the final decision to recommend zipcodes to our clients.
Linear Regression - Predicting Student Debt
I researched on student data for one of my statistical coursework projects. The dataset consisted of a random sample from the population of recent graduates that lived in Allendale during their time at GVSU. The main goal of this project is to choose the best model to be able to predict the amount of debt to be paid by students. For more explanation of the project, please read this article:
Tic Tac Toe using Reinforcement Learning
In this project, I have tried out the concepts of reinforcement learning by applying Q-learning to solve Tic-Tac-Toe game. Two computer agents were trained to be able to evaluate tic-tac-toe positions in order to return the best move by playing against each other for many games.