🧠 Sentiment Classification & 🧬 Medical Image Generation with Deep Learning

This repository contains two mini projects showcasing practical applications of deep learning in Natural Language Processing (NLP) and Computer Vision (CV):

1. IMDb Sentiment Analysis using Word2Vec + RNN variants
2. Medical Image Generation using Deep Convolutional GAN (DCGAN)

---

📝 Table of Contents

• Project 1: IMDb Sentiment Analysis
• Project 2: DCGAN on PathMNIST
• Sample Outputs
• Key Learnings

---

📌 Project 1: IMDb Sentiment Analysis

Dataset: IMDb Movie Reviews (Kaggle)
Goal: Classify movie reviews into positive or negative sentiments.

🔨 Steps:

• Preprocessed text: tokenization, stopword removal, lemmatization, HTML tag removal, lowercasing.
• Used Word2Vec embeddings to convert text to vectors.
• Built multiple deep learning models:
  • Simple RNN
  • GRU
  • LSTM
  • BiLSTM

📊 Results:

Model	Accuracy	Epochs
BiLSTM + Word2Vec	71.5%	9
LSTM + Word2Vec	69.5%	23
GRU + Word2Vec	54.1%	15
RNN + Word2Vec	50.5%	20

---

📌 Project 2: DCGAN on PathMNIST

Dataset: PathMNIST (MedMNIST)
Goal: Generate realistic synthetic pathology images using DCGAN.

🔨 Architecture:

• Generator: Deconvolution layers to upsample noise into images.
• Discriminator: CNN layers to distinguish real from fake images.
• Trained for 1000 epochs, monitoring Generator/Discriminator loss and FID score.

📉 Performance:

• Final FID Score: 220.91
• Best FID: 187.17 (at epoch 300)
• Loss curve analysis indicated mode collapse:
  • Generator output became repetitive.
  • Discriminator became too confident.

📁 Refer to imdb_sentiment_analysis_and_DCGAN.ipynb

---