Using data mining techniques for predicting individual tree mortality in tropical rain forest: logistic regression and decision trees approach

  • Authors:

  • Yasmin Yahya;Roslan Ismail;Samreth Vanna;Khorn Saret

  • Affiliations:

  • Universiti Kuala Lumpur, Kuala Lumpur, Malaysia;Universiti Kuala Lumpur, Kuala Lumpur, Malaysia;Phnom Penh, Cambodia;Phnom Penh, Cambodia

  • Venue:
  • Proceedings of the 8th International Conference on Ubiquitous Information Management and Communication
  • Year:
  • 2014

Quantified Score

Hi-index 0.00

Visualization

Abstract

Tree mortality is an essential process in forest ecosystem dynamics. It is one of the least understood phenomena due to species-rich in tropical rain forests. Individual tree mortality model was developed for predicting the probability of mortality in dipterocarpaceae tree family group in Koh Kong province, Cambodia. Finding appropriate methods for modeling mortality have often proved to be a difficult challenge. Two data mining methods were performed in this study; logistic regression and decision trees. Chi-squared Automatic Interaction Detector (CHAID) method was chosen for decision trees method since it always chooses the independent variable that has the strongest relation with the dependent variable. The probability of mortality decreased with increasing individual tree basal area. The performance of each model from both methods was compared using calibration (chi-test) and discrimination (area under receiver-operating characteristic curve or c-index). The study presented that logistic regression outperformed decision trees for both calibration and discrimination. The model developed is expected to improve the accuracy of the stand forecast.