[Paper Review] Explainable AI Integrated Feature Engineering for Wildfire Prediction

Di Fan, Ayan Biswas+1 more|arXiv (Cornell University)|Apr 1, 2024

Fire effects on ecosystemsEnvironmental Science3 citations

TL;DR

This study proposes an explainable AI-integrated machine learning framework for wildfire prediction, combining XGBoost for classification and Random Forest for regression, with a hybrid CNN for joint classification and regression using numerical and image data. Key results show XGBoost achieved high accuracy in classifying wildfire types, while Random Forest regression minimized error in predicting burned area, and XAI techniques like TreeSHAP and Grad-CAM revealed that soil moisture and ignition conditions are the most influential predictors.

ABSTRACT

Wildfires present intricate challenges for prediction, necessitating the use of sophisticated machine learning techniques for effective modeling\cite{jain2020review}. In our research, we conducted a thorough assessment of various machine learning algorithms for both classification and regression tasks relevant to predicting wildfires. We found that for classifying different types or stages of wildfires, the XGBoost model outperformed others in terms of accuracy and robustness. Meanwhile, the Random Forest regression model showed superior results in predicting the extent of wildfire-affected areas, excelling in both prediction error and explained variance. Additionally, we developed a hybrid neural network model that integrates numerical data and image information for simultaneous classification and regression. To gain deeper insights into the decision-making processes of these models and identify key contributing features, we utilized eXplainable Artificial Intelligence (XAI) techniques, including TreeSHAP, LIME, Partial Dependence Plots (PDP), and Gradient-weighted Class Activation Mapping (Grad-CAM). These interpretability tools shed light on the significance and interplay of various features, highlighting the complex factors influencing wildfire predictions. Our study not only demonstrates the effectiveness of specific machine learning models in wildfire-related tasks but also underscores the critical role of model transparency and interpretability in environmental science applications.

Motivation & Objective

• To improve wildfire prediction accuracy by integrating machine learning models tailored for classification and regression tasks.
• To enhance model interpretability in environmental applications using explainable AI (XAI) techniques.
• To identify the most influential environmental and meteorological features driving wildfire outcomes through feature importance analysis.
• To develop a hybrid deep learning model that processes both numerical data and satellite imagery for simultaneous classification and regression.
• To validate model performance using real-world wildfire datasets from public repositories.

Proposed method

• Employed XGBoost for wildfire type classification and evaluated performance using accuracy and robustness metrics.
• Used Random Forest regression to predict burned area extent, optimizing for prediction error and explained variance.
• Developed a hybrid CNN model that processes both meteorological data and satellite imagery for joint classification and regression tasks.
• Applied TreeSHAP to quantify feature importance and model output contributions in regression predictions.
• Used LIME to assess local model behavior and linearity around individual predictions for inside and outside fire areas.
• Utilized Partial Dependence Plots (PDP) to visualize the marginal effect of features on predicted fire extent.

Experimental results

Research questions

• RQ1 Which machine learning model performs best for classifying wildfire types or stages?
• RQ2 How accurately can regression models predict the extent of burned areas in wildfire events?
• RQ3 Which environmental and meteorological features most significantly influence wildfire prediction outcomes?
• RQ4 How do XAI techniques like TreeSHAP, LIME, and PDP reveal the decision-making logic of wildfire prediction models?
• RQ5 Can a single CNN model effectively perform both classification and regression on wildfire data using image and numerical inputs?

Key findings

• XGBoost achieved the highest accuracy and robustness in classifying wildfire types or stages, outperforming other models in classification tasks.
• Random Forest regression demonstrated superior performance in predicting burned area, with the lowest prediction error and highest explained variance.
• For inside fire area prediction, soil moisture (smois) and ignition conditions had the strongest positive impact, as revealed by TreeSHAP analysis.
• For outside fire area prediction, soil moisture (smois) and wind speed were the most influential features, with wind speed showing significant impact at values over 10.
• The hybrid CNN model achieved a 93.10% classification accuracy and regression errors of 0.75% (inside) and 2.41% (outside) burned area.
• Grad-CAM visualization confirmed that the model focused on critical regions in satellite images, such as burn scars and fuel patterns, to inform classification decisions.