Flood Prediction Alert System in the state of Odisha, India: An Explainable and Actionable AI based predictive model

Abstract

Floods are a recurrent and devastating natural disasters, cause immense loss of life, infrastructure damage, and economic hardship. Mitigating these impacts requires accurate and timely flood prediction. Traditional hydrological models, while valuable, often rely on complex physics and extensive data. This thesis explores the application of explainable machine learning (XAI) for flood prediction, offering a data-driven alternative with interpretable results. Black-box machine learning models excel at prediction but lack transparency. XML techniques bridge this gap, enabling us to understand the model's decision-making process. This is crucial in flood prediction, where understanding the factors contributing to a predicted flood is critical for effective risk mitigation strategies.
Thesis Key Objectives:
I.
To develop an explainable machine learning model for flood prediction. This model will utilize historical data on factors influencing floods, such as rainfall patterns, river discharge measurements, land use data, and digital elevation models (DEMs).
II.
To evaluate the model's accuracy and explainability. The model's performance will be assessed using standard flood prediction metrics like Root Mean Squared Error (RMSE) and Nash-Sutcliffe Efficiency (NSE). XML techniques like SHAP (SHapley Additive exPlanations) will be employed to understand feature importance and gain insights into the model's predictions

III.
To compare the XAI model with traditional approaches. This will involve comparing the prediction accuracy, explainability, and computational efficiency of the XAI model with existing hydrological models.
IV.
To open up new avenues for further research into this space. The idea is also to open new opportunities and avenues in this space of Flood Prediction using AI and with each passing day creating newer horizons for AI and generative AI, probably this research will open and inspire new researchers to come up with new, modern and more sophisticated techniques in future.
Expected Benefits:
I.
Enhanced Flood Prediction Accuracy: The XML model is designed to deliver precise flood forecasts, providing critical lead time for disaster response and preparedness.
II.
Improved Decision-Making Capabilities: By elucidating the key factors influencing flood forecasts, stakeholders can devise focused mitigation plans, ensuring the protection of at-risk areas and essential infrastructure.
III.
Greater Transparency and Trust: XML promotes confidence in the model's forecasts by offering interpretable results, enabling stakeholders to make well-informed decisions based on transparent data.