Research / intelligence lab

Machine intelligence for a changing climate.

I explore explainable AI, climate-risk modelling, rare-event forecasting, and energy-aware machine learning through reproducible student research.

Three completed computational pilot studies · September 2026
Selected studies / 01

Research manuscripts

Public data · Reproducible experiments · Transparent limitations

01

Explainable AI for Monthly Climate Stress Early Warning

A NASA POWER pilot across eight Bangladesh divisional cities with temporal validation and permutation-based explanations.

0.923ROC-AUC
0.821PR-AUC
Explainable AIClimate riskRandom Forest
Read PDF
02

Machine Learning for Next-Day Extreme Rainfall

A multi-city rare-event study using chronological testing and a rainfall-based flood-risk proxy.

78.1%Recall
0.348PR-AUC
Machine learningRainfallEarly warning
Read PDF
03

Carbon-Aware Machine Learning Model Selection

An accuracy-energy benchmark across three public datasets with an explicit carbon-aware decision rule.

93.9%Energy reduction
0.31 ppAccuracy cost
Green AIEfficiencyBenchmarking
Read PDF
Research status. These are completed computational pilot studies and have not been peer reviewed or accepted by a journal or conference. Reported values come from the documented experiments; each paper states its data and measurement limitations.
Reproducibility / 02

Follow the evidence.

The research directory contains the downloadable manuscripts. Experiment methods, results, and limitations are documented in each paper.

Research files