Projects

Machine Learning

HybridLSTM for NIFTY predictions

Used long short-term memory networks to conduct 3 experiments for predicting future prices of the NIFTY50 index using 20 technical indicators, 13 global indices and a combination of both. The performance was evaluated on MSE, MAE and R2 score.

COVID Multiclassifier

A machine learning model which uses Siamese network based architecture to perform multi-class classification to identify and separate Pneumonia, Normal, COVID-19, and Severe COVID-19 patients on the basis of the X-Ray images. The obtained accuracy is 98%.

Code's cooking, check back soon.

Movie Reviews Sentiment Analysis

A multilayer perceptron model to classify the movie reviews from the IMDB dataset as positive or negative. The obtained accuracy on test set is 85%.

Imagenet4U

Implementation of deep CNN to classify horses, dogs, humans and cats trained on a part of the imagenet dataset. Training and validation generator used in pipeline to accomodate large data.

Emoji Predictor

Implementation of sequence models, here LSTM RNN to predict emoji for a given line of text. The achieved accuracy is 76%.

Machine Learning Mini Projects

  1. Real-time face recognition

A project which uses the K-Nearest Neighbours algorithm to recognise single/multiple faces from the live video frames.

  1. Analysis of MNIST Dataset

Implementation of a small size convolutional neural network using Keras API and Naive Bayes classifier for recognition of handwritten digits using the MNIST dataset. The obtained accuracy is 99%.

  2. Analysis of fashion MNIST Dataset

Convolutional Neural Networks implementation using Keras API for analysis on the fashion MNIST dataset. The obtained accuracy on the test set is 86%.

  2. Titanic Survivor Prediction Challenge (Kaggle)

Implementation of decision tree classifier from scratch and comparison against sklearn implementation of decision trees and random forest on Kaggle's Titanic Survivor Prediction dataset. The obtained accuracy is 86%.

  2. Housing Prices Prediction (Kaggle): The analysis of performance of linear regression against a neural network on the house prices dataset of Kaggle challenge.

Robotics

CraSOS

A standalone device with the capabilities of accident prevention, detection and recovery. On collision detection, SOS message(containing location, medical info) generated and sent to helpline numbers, 3 chosen emergency contact numbers and broadcasted using BLE Beacon & nearby notifications.

e-Yantra Robotics Competition 2019-20

Theme: Rapid Rescuer

Code's cooking, check back soon.

Lane Detection App

This project uses computer vision techniques and Hough transform to perform fast and light-weight lane detection for autonomous vehicles.