CIS 6930 - Applied Machine Learning Using Python
1 Input Dataset
We will use the UCI Mushroom (https://archive.ics.uci.edu/ml/datasets/ mushroom) datasets for this project. This dataset contains about 8124 mushroom examples over 22 categorical attributes, with each example labeled either edible (“e”) or poisonous (“p”). In this dataset, there are 4208 (51.8%) edible examples and 3916 (48.2%)poisonous examples. For your experiments once the decision tree model is implemented, this dataset is randomly split to a training set of 80% and a testing set of 20%. Both have very similar distribution: 51.9% edible and 48.1% poisonous for the training set, and 51.4% edible and 48.6% poisonous for the testing set. Both training.data and testing.data can be downloaded from Canvas and will be used for this project.
2 Requirements
1. The program should implement Quinlan’s ID3 and C4.5 decision tree learning algorithms using the template shown in page 5 of slides dt.pdf, where the IMPORTANCE method would be based on information gain and gain ratio, respectively.
2. For both ID3 and C4.5, the program should implement a 10-fold cross validation process over the training set, pick the best model with the highest F1 score, and run it finally on the testing set and report the F1 score.
3. Based on the results, compare ID3 and C4.5.
4. Note that, to ensure non-randomness, the 10 partitions used in cross validation are provided as 10 files for you to download, i.e., training aa.data through training aj.data.
