Wildfire Predictions: Determining Reliable Models using Fused Dataset

Hariharan Naganathan, Sudarshan P. Seshasayee, Jonghoon Kim, Wai K. Chong, Jui-Sheng Chou

Research output: Contribution to journalArticlepeer-review

Abstract

Wildfires are a major environmental hazard that causes fatalities greater than structural fire and other disasters. Computerized models have increased the possibilities of predictions that enhanced the firefighting capabilities in U.S. While predictive models are faster and accurate, it is still important to identify the right model for the data type analyzed. The paper aims at understanding the reliability of three predictive methods using fused dataset. Performances of these methods (Support Vector Machine, K-Nearest Neighbors, and decision tree models) are evaluated using binary and multiclass classifications that predict wildfire occurrence and its severity. Data extracted from meteorological database, and U.S fire database are utilized to understand the accuracy of these models that enhances the discussion on using right model for dataset based on their size. The findings of the paper include SVM as the best optimum models for binary and multiclass classifications on the selected fused dataset.

Original languageAmerican English
Pages (from-to)28-40
JournalGlobal Journal of Computer Science and Technology: C Software & Data Engineering
Volume16
Issue number4
StatePublished - Jan 1 2016

Keywords

  • binary and multiclass classifiers
  • decision tree stumps
  • forest fire
  • k-fold cross-validation
  • k-nearest neighbor
  • support vector machines

Disciplines

  • Civil Engineering
  • Construction Engineering and Management

Cite this