Post-Hoc Interpretability in Time-Series Classification

Project by Balaji Viswanathan, Sana Begum, Prajuvin Prabha, Sarim Ali, and Kausar Ali Ansari.

---

Project Overview

In safety-critical domains (healthcare, finance, etc.), model interpretability is as important as accuracy. We explore whether post-hoc attribution methods can reliably explain and validate deep-learning models on a benchmark time-series dataset.

• Dataset: Banknote Authentication (treated here as an ECG-style 1D time series)
• Models:
  • 1D Convolutional Neural Network (CNN)
  • Long Short-Term Memory network (LSTM)
• Attribution Methods:
  1. Integrated Gradients
  2. DeepLift
  3. GradientShap
  4. KernelShap

---

Installation & Usage

1. Clone this repo

   git clone https://github.com/YourUsername/TimeSeries-Interpretability.git
   cd TimeSeries-Interpretability