Chapter 14 - Fuzzy-machine learning models for the prediction of fire outbreaks: A comparative analysis

Abstract

This chapter compares fuzzy-machine learning algorithms for predicting fire outbreaks using temperature, smoke, and flame datasets. The datasets are preprocessed using interval type-2 fuzzy logic (IT2FL). Min-max normalization and principal component analysis (PCA) are used to predict, normalize, and select relevant feature labels in the dataset. The preprocessed datasets are used to train (80%) and test (20%) K-nearest neighbor (KNN), support vector machine (SVM), random forest (RF), linear discriminant analysis (LDA), and classification and regression tree (CART). K-fold cross-validation is used to evaluate the performance of the models using the receiver operating curve (ROC), specificity and sensitivity matrices. The validation result shows that KNN performs better with ROC values of 0.99878 against 0.99753, 0.997265, 0.997073, and 0.958693 for SVM, RF, LDA, and CART, respectively. It is also observed that KNN outperforms SVM, RF, LDA, and CART in predicting fire outbreaks with the highest degree of accuracy of 0.9643 against 0.9571, 0.95, 0.9429, and 0.9571, respectively.