Aspiring Data Scientist

github

Narender Kumar  • Peerlist

–Collaborated on a Pix2Pix GAN image restoration project using U-Net Generator and PatchGAN Discriminator. –Web-scraped 54,380 images via Selenium and applied Gaussian noise and motion blur for dataset corruption. –Attained results: SSIM 0.9235, PSNR 28.69, LPIPS 0.0979, with 0.18 ms/image inference –Designed an interactive Streamlit web application for real-time restoration of corrupted images to clean versions

Clear Vision

–Built and implemented a custom CNN model for multi-disease classification of Chest X-Ray images. –Enhanced model robustness through Chest X-Ray specific data augmentation and L2 regularization, while employing focal loss to mitigate class imbalance. –Attained F1: 0.92 & Recall: 0.95 and F1: 0.91 & Recall: 0.94 for Tuberculosis and Pneumonia respectively. –Applied Grad-CAM heatmaps to demonstrate model alignment with radiological expertise, increasing trust in AIdriven diagnosis.

Explainable AI for Multi-Class Chest X-Ray Di

-Fetched historical data of 16 Indian banking stocks using yfinance for next-day price prediction. –Applied per-ticker MinMax scaling and log-transformation on volume to handle scale differences and skewness. -Optimized hyperparameters using Optuna for LSTM, GRU, XGBoost, LightGBM and CatBoost and trained models with tuned configurations. -Achieved highest performance with LightGBM, obtaining an R2 score of 0.93 on test data.