My Personal Projects
Cataract Detection Project
This project implements a robust eye/cataract detection system using deep learning techniques. The system leverages YOLOv8 for object detection to identify and classify eye conditions, particularly focusing on cataract detection. The system automatically analyzes and provides comprehensive dataset statistics.
Debris Accumulation Forecasting
The system provides marine debris accumulation forecasting using a combination of time series models, machine learning algorithms, and feature engineering. It analyzes historical debris data from NASA's marine debris dataset to forecast future accumulation trends, enabling better resource allocation for cleanup operations and environmental monitoring.
Marine Debris Classification
A comprehensive ML pipeline that analyzes environmental data to predict debris sources and pollution severity levels. The system analyzes environmental data, debris characteristics, and geographic information to predict Primary debris source, Debris material classification and Pollution severity levels.
Mental Health Chatbot Project
A Full stack web project with Python backend and NuxtJS frontend. Uses transfer learning to make an LLM from the DiabloGPT model. Stored as two separate repositories on GitHub. This card leads to the backend repository. Explore my GitHub account for the front end repository.
Prescription Project
A dummy data on users' prescriptions and their filling history containing about 250,000 entries is generated with the help of ChatGPT. This data is then trained through various models like Decision Tree, GBM, Neural Networks and more to compare the reliability of the data as proof of concept on each model
Equipment Failure Project
A historical weather data is downloaded from SMHI. With the help of ChatGPT and further research, a dummy data is generated based on weather data to simulate virtual failure of open equipments such as mobile towers. This data will be trained on several models to compare the results
Marine Audio Detection
A limited data is obtained from a public library containing noisy voices of dolphins, whales, ships and more. A model is created and the data is trained on the same model after processing through various methods like Spectrogram, Mel Hamming and such to compare results as a comparison of pre-processing methods
Sign Language Prediction
Data is Generated through using webcam and creating your own data. It is then labelled and processed. A CNN model is trained to predict on the generated data and the trained model is saved. This saved file is then loaded on another python file that accesses the webcam and predicts the signs shown on the camera within the ROI.
Crypto-Currency Stock Prediction
An open dataset from kaggle is used to train an LSTM model to predict values after the next few minutes. The model is saved after training and another python file loads this model. It gets live crypto-currency values from an open website and predicts expected values every 5 minutes. This interval can be updated in the python file.
Covid Hotspot Prediction
Covid data is scrapped from Kerala Government's live website using a scraping code and Chromium extension. Data is categorized into districts. This data is used to train a Random Forest Classifier Model and the model is saved. It is then used to predict next day's covid positive count based on today's statistics and input district.
Stock Price Prediction
The NSE Tata dataset available freely online is downloaded and used to create an LSTM model that predicts the next stock closing price for a given firm with previous day's statistics. The model is saved after training but not implemented into a complete ecosystem since it was one of my earlier works.