Urban Land Cover Data Analysis

The dataset was download from https://archive.ics.uci.edu/ml/datasets/Urban+Land+Cover.

For the training set, there are 168 instances with 148 attributes.

Abstract

The primary objective of this project is to conduct a comparative analysis of various high-dimensional feature selection techniques on an Urban Land Cover Data Set. The techniques under scrutiny include Principle Component Analysis, regularization by thresholding, and penalization. By interchanging the training set and testing set, the dimension of the problem remains high, thus enabling an evaluation of the impact of instances on the same distribution. The findings suggest that the thresholding method improves the performance of Principle Component Analysis, whereas $l_1$ penalization demonstrates greater generality in handling such datasets.