[Solved] CSE455-Homework 1

Experiments with KNN, SVM and DT on a well known classification data set (Leaf data. You can use various Python libraries to download these data files.

This assignment is expecting you to write five different functions to test your solutions to the problem. You are expected to use the Python language. You will prepare a Jupyter Notebook including your code and results.

• Part 1: Build a classifier based on KNN (K=5 for testing) using Euclidean distance.
  • You are expected to code the KNN classifier by yourself.
  • Report performance using an appropriate k-fold cross validation using confusion matrices on both datasets.
  • Report the run time performance of your above tests.
• Part 2: Build a classifier based on KNN (K=5 for testing) using Manhattan distance.
  • You are expected to code the KNN classifier by yourself.
  • Report performance using an appropriate k-fold cross validation using confusion matrices on both datasets.
  • Report the run time performance of your above tests.
• Part 3: Build a classifier based on linear SVM.
  • You may use an available implementation of SVM in Python.
  • Report performance using an appropriate k-fold cross validation using ROC curves and confusion matrices. Find the best threshold for the SVM output as described in the note by Fawcett.
  • Report the run time performance of your above tests.
• Part 4: Build a classifier based on polynomial SVM.
  • You may use an available implementation of SVM in Python.
  • Report performance using an appropriate k-fold cross validation using ROC curves and confusion matrices. Find the best threshold for the SVM output as described in the note by Fawcett.
  • Report the run time performance of your above tests.
• Part 5: Build a classifier based on DT (Decision Trees).
  • You may use an available implementation of DTs in Python. o Use different pruning strategies.
  • Report performance using an appropriate k-fold cross validation. o Report the run time performance of your above tests.
  • Write a function to convert one of your decision trees into a set of rules (i.e., extract the path to each leaf nodes).
• Part 6 (optional): Improve your search procedure in Part 1 and Part 2 using an advanced search algorithm such as kd-trees.