publications

2026

1. Review paper

   

   Mapping the latency explainability accuracy Pareto frontier : a systematic benchmark and framework for attribution methods in real time satellite anomaly detection

   Ekaagra Gupta

   Targeting IEEE Transactions on Geoscience and Remote Sensing, 2026

   Abs

   Satellite anomaly detection systems require accurate and computationally efficient attribution methods. This paper introduces a benchmarking framework that jointly evaluates latency, explainability quality, and predictive performance across multiple attribution algorithms. We construct a Pareto frontier to identify practical trade-offs for real-time deployment on edge and cloud GPU systems.

2. Patent

   

   GETHER: A Generative Emission Temporal Hybrid Explainable Regression Framework

   Ekaagra Gupta

   Patent under examination, 2026

   Abs

   Developed an end-to-end temporal deep learning pipeline for Air Quality Index (AQI) forecasting using a stacked LSTM architecture (128→64 units) with engineered lag and rolling-statistic features, achieving improved predictive performance over Linear Regression and Random Forest baselines. Integrated a causal discovery module leveraging Granger causality and dynamic causal graphs to uncover temporal relationships between key pollutants (PM2.5, NO₂, and SO₂) and AQI fluctuations. Enhanced model transparency through a SHAP-based explainability framework that quantified global feature importance and interpreted individual predictions. Additionally, designed a counterfactual policy simulation module to estimate AQI changes under hypothetical emission-reduction scenarios, and deployed the complete system through an interactive Streamlit dashboard for real-time analysis and decision support.