Synthetic_Stock_Data

🧠 Synthetic Stock Data Generator & Visualizer

This project builds a synthetic stock market data generator using a combination of Autoencoders (AE) and Generative Adversarial Networks (GANs).
The goal is to create realistic synthetic financial time-series data and compare model performance between real and synthetic datasets.

---

🚀 Project Overview

🔹 Workflow

1. Autoencoder (AE):

   • Encodes stock price data into a compressed latent space.
   • Captures temporal and feature-based dependencies between Open, High, Low, Close, and Volume.

2. GAN (Generator + Discriminator):

   • Learns to generate synthetic latent vectors that mimic the AE latent representations.
   • Generator produces fake latent vectors.
   • Discriminator learns to distinguish between real (from AE encoder) and fake (from Generator).

3. Synthetic Data Reconstruction:

   • The synthetic latent vectors are passed through the AE Decoder.
   • This recreates synthetic stock market data at the feature level (Open, High, Low, Close, Volume).

4. Model Evaluation:

   • A downstream neural network classifier is trained on:
     • Real data
     • Synthetic data
   • Performance metrics and comparison charts are saved in the /charts folder.

---

📊 Visualization App

The project includes a Gradio-powered dashboard to visualize stock time series for real and synthetic data.

🖥️ Try it on Hugging Face

If you’re viewing this on Hugging Face, launch the app directly below 👇

🔍 App Features

• Select any stock ticker and feature (Open, High, Low, Close, Volume)
• View 5-year time series comparisons of original vs synthetic data
• Interactive plots rendered with matplotlib

---

📂 Repository Structure