Feature Selection Using Fast Correlation-Based Filter (FCBF) For Leaf Classification

Lua, Xin Lin (2015) Feature Selection Using Fast Correlation-Based Filter (FCBF) For Leaf Classification. Project Report. UTeM, Melaka, Malaysia. (Submitted)

Text (24 Pages) Feature Selection Using Fast Correlation-Based Filter (FCBF) For Leaf Classification 24 Pages.pdf - Submitted Version Download (267kB)

Abstract

Imagine taking a forest trail hike, where you see many interesting plants. How would you differentiate one plant from another based on its leaves? Do you base your decision on its color? Or its texture? Or size? The thing is, one single leaf could give you a lot of information, from color to texture to size, so on and so forth. But the problem is: which features to focus on so as to make a good decision? The identification of a feature subset that best represents a class so as to build a strong predictive model is still an issue that researchers are still working on solving. This paper focuses on feature selection, where the Fast Correlation-Based Filter is compared to the Correlation-based Feature Selector. Four datasets were retrieved from the UCI Machine Learning Repository. The SVM classifier was used to build a predictive model based on the feature subset. The experiments were done using the weka machine learning tool, with the evaluation criterion being predictive accuracy, kappa measure, and time taken to build a model.

Actions (login required)

View Item