Predicting fire incidents with ML: an XAI approach

Abstract

With increasing urbanization and industrial activities in Dublin city, there is a growing interest in natural disasters and accidents. In this research, we compared the regression performances for the number of fire incidents in Dublin with seven different machine learning models. We evaluated how each model performs with metrics like Mean Squared Error (MSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and R-Squared scores. Among the used models which are Prophet, Auto-Regressive Integrated Moving Average (ARIMA), Simple Linear Regression, Polynomial Regression, Support Vector Regression, Decision Tree Regression, and Random Forest Regression, the most precise model was the Prophet with the highest R-Squared score of 0.91 because it effectively captures the underlying trend, seasonality and holiday effects. This research also aims not only to establish the superior model but also to give clear and understandable reasons for these predictions using explainable AI (XAI). In particular, the trust in Prophet is enhanced by global explanation and local explanation for users to believe in the decision-making processes in the model. It enabled us to enhance the interpretability and transparency of the Prophet, which is aligning with ethical AI (Artificial Intelligence).