Research focused on understandable and useful AI.
My research interests sit at the intersection of applied machine learning, multimodal evidence, scientific forecasting and explainable decision support.
Peer-reviewed Springer chapter.
A Deep Learning-Based Study for Cyclone Track Forecasting
Majid, R., Hasan, A., Mostarin, S., Alam, K.R., & Rahman, R.M. (2024). "A Deep Learning-Based Study for Cyclone Track Forecasting: Comparative Analysis Using Historical Data from the Bay of Bengal." In Networking and Parallel/Distributed Computing Systems, Studies in Computational Intelligence, Vol. 1125. Springer, Cham.
The study compared recurrent deep-learning architectures for predicting cyclone trajectories using historical Bay of Bengal data. Performance was evaluated using both statistical metrics and geographical distance error so that model behaviour could be interpreted in a practical forecasting context.
Cyclone-track forecasting supports disaster preparation. Using kilometre-based distance error makes recurrent-model outputs interpretable in an operational forecasting context.
- Operational forecasting
- Predictive maintenance
- Movement modelling
- Sequential prediction
LenseESG — explainable multimodal ESG risk.
Can structured environmental, social, governance and controversy indicators be combined with company disclosures and news-derived text to produce a more informative and explainable ESG risk-assessment workflow?
The multimodal classification experiment achieved approximately 0.71 weighted F1, outperforming the weaker text-only approach. SHAP and transformer-attention analysis surfaced influential risk variables and textual signals; leakage-prone configurations were identified and corrected.
Limitations: ESG labels are provider-dependent; news coverage is uneven; a 542-company dataset limits generalisation; attention is not a complete explanation of model reasoning.
Where I focus my attention.
- Explainable and trustworthy AI
- Multimodal machine learning
- NLP and domain-specific language models
- Scientific and time-series machine learning
- Responsible AI evaluation
- Human-centred decision support
- AI for risk and sustainability analysis
How I think about evidence.
Evidence before claims
Results should be connected to transparent data, evaluation and limitations.
Explanations for decisions
Explainability should help a user inspect, challenge or act on a result.
Models in context
Technical performance must be considered alongside domain meaning and operational use.
Honest uncertainty
Research communication should distinguish evidence, association, limitation and inference.