Explaninable AI

Navigation

• HomePage
• Project Details
  • Introduction
  • Stock Market
  • Prediction Result
• Technical Methodology
  • Gramian Angular Field
  • FastAI CNN Model
  • Grad-CAM Algorithm

What did we do?

Source from: Google Image, Google Image

While most prevalent approaches to stock prediction might base around modelling with time series data, this research purposefully structures this attempt as an image classification task – both to explore deep learning algorithms’ capability on learning non-conventional images, and to inspect explainability of these neural networks.

How did we do it?

Prediction Model

Source from: Google Image

1. Encode time series data as images via Gramian Angular Field.
   
2. Train CNN model for pattern learning.
   
3. Apply Grad-CAM algorithm over the learned CNN model for visual explanations.