Imagery Data Impact Analysis in Machine Learning Algorithm Performance for Bushfire Detection Systems

The increased number and intensity of recent bushfire events has become a critical environmental and economic issue. A robust bushfire detection mechanism is crucial for mitigating their impact. Machine learning (ML) and deep learning (DL) algorithms have been extensively employed in this context. However, understanding the factors influencing bushfire detection can enhance algorithmic performance. This study assesses the impact of diverse datasets on ML and DL algorithms efficiency and effectiveness in detecting bushfires across three ML algorithms (support vector machine (SVM), logistic regression (LR), and random forest (RF)) and three CNN algorithms (VGG16, VGG19, and ResNet50). Datasets of varying quantity and quality were utilized to evaluate algorithm performance. Our primary objective is to identify influential factors related to data quality and quantity for imagery analysis in bushfire detection and recommend effective algorithms. Our findings reveal that the optimal detection algorithm depends on factors such as specific goals, dataset size, computational resources, and trade-offs between accuracy, interpretability, and complexity. Combining different algorithms through techniques like model stacking or ensembling can offer additional value.