Principal Component Analysis and Self-Organizing Map for Visualizing and Classifying Fire Risks in Forest Regions

概要

論文の詳細を見る
Dataset compiled from spreading hot spots, responsible for fire risk in many regions of Indonesian forests, are complex, primarily induced by the large size of the observed regions and high variation of hot spot distribution. The challenge in analyzing this type of dataset is to develop statistical techniques that facilitate the analysis, visualization, and interpretation of the results. Techniques, such as multivariate analysis and artificial neural networks, have been applied to resolve the high-dimensional space in such large datasets. Each method uses a different rationale for how the relationship between the input parameters will be preserved during analysis. This study presents the use of a principal component analysis (PCA) and a self-organizing map (SOM) to reduce the high dimensionality of the input variables and, subsequently to visualize the dataset into a two-dimensional (2-D) space. The results indicate that the first two principal components of the PCA provide a large percentage of cumulative variance to explain the data patterns. However, a comparison of the data projection, SOM is better suited than PCA in visualizing the fire-risk distribution in forests. The SOM color-coding and labeling also effectively visualized a classification system of fire risk via node clusters, in such a way that the fire risks level according to their hot spot locations in forest is easily interpreted.
農業情報学会の論文