XAI Climate Identifier
A climate-prediction model paired with explainable AI, designed to make both a long-term forecast and the reasoning behind it easier to understand through a simple web interface.
- Role
- Lead Implementer
- Collaborators
- T Jathin · S Vishnu
- Duration
- 7 weeks
- Tools
- Python
Model
The model uses polynomial regression to estimate long-term climate outcomes. After training and testing, it reached 74.567% accuracy on the project dataset and was integrated into a browser-based product.
Explanation layer
The workflow keeps preprocessing reproducible and exposes model outputs for interpretation. Its explanation toolkit includes saliency maps, Grad-CAM, integrated gradients, occlusion, and SmoothGrad, with heatmaps and comparison views that make attribution patterns visible.
Evaluation
Explanations were considered through faithfulness, robustness, complexity, localization, and model-randomization sanity checks. The aim was to test whether an explanation reflects model behavior and remains stable under noise and input transformations.
Data constraint
The original plan called for a small, region-specific climate dataset. Because suitable regional data lacked the required spatial and temporal consistency, the project moved to broader, more accessible climate sources.