Integrating Global and Local Explainability in Machine Learning and Deep Learning for Drought Forecasting of Bangladesh
Students & Supervisors
Student Authors
Supervisors
Abstract
"Drought represents a major climate hazard to Bangladesh resulting in serious threats to agricultural productivity, water availability, and food security. This study proposes the introduction of XAI (Explainable Artificial Intelligence) techniques for predicting drought through the integration of advanced machine learning and deep learning algorithms with modelling interpretability techniques at both the global and local levels. The use of NASA POWER climate projections (2012-2024) for 38 districts demonstrated that regression analysis produced a very high degree of predictive fit of (R2 = 0.953) for predicting the amount of moisture in soils. The classification benchmarking results showed that Random Forest (RF) and Extra Trees (ET) had the highest predictive performance (approximately ≈ 98%), with a balanced precision/recall (positive and negative) across drought vs non-drought classes, while LSTM and GRU also performed well, but were less accurate. Through SHAP analysis, relative humidity, dew point, and surface temperature were identified as being the most important predictors of a drought event; LIME allowed for a case-by-case interpretation of the XAI model’s predictions to provide decision support and improve transparency. By combining both high levels of predictive accuracy, along with an explainable model, the proposed framework provides reasonably accurate and interpretable drought forecasting to improve agricultural planning, water management, and climate adaptation efforts in regions that are highly vulnerable to drought."
Keywords
Publication Details
- Type of Publication:
- Conference Name: 28th International Conference on Computer and Information Technology (ICCIT)
- Date of Conference: 19/12/2025 - 19/12/2025
- Venue: Long Beach Hotel, Cox's Bazar, Bangladesh
- Organizer: IEEE Bangladesh Section