CPSC 483 - Data Mining and Pattern Recognition - Implementation of Machine Learning Algorithms

Team: Phillip Ly, Holly Ho, Sara Lipowsky, Tony Dao, James Lindsey

Link to report and PPT slides: https://goo.gl/ab6WQy

The goal is to train, validate, and test Multi-Layer Perceptrons or RBF Networks by implementing classification and regression (including time series prediction over a period of 8000 hours) algorithms on Heart Disease and Air Quality data sets from UC Irvine Machine Learning Repository. We've attained an accuracy as high as 95% for classification of heart disease.

Steps to Replicate Our Results

Prerequisites:

• Python 2.7
• numPy
• matplotLib

1. Download project from GitHub repository:

• In terminal type: git clone https://github.com/PhillipLy/CS483.git
• Notice there are 2 folders- Heart and Air.

For Classification:

2. Change into “Heart” directory:

• cd Heart
• python heart.py
• The range of % that we got
• The range is 73% - 94.5%
• Changes to the code from the original
• Used a different dataset (called reprocessed.hungarian.data)
• Created matrices that allowed access of all usable features
• Modified the test matrix to fit the output of our algorithm (only 0 or 1 instead of 0-2)
• Made global variables so that it is easier to test different features
• Changed the number of hidden nodes from 5 to 3

For Regression:

3. Change into “Air” directory

• cd Air
• python air.py
• The range of % that we got
• The Error Range: 22% - 26.67% with t = 3 and k = 10
• Changes to the code from the original
• Used reprocess data (called AirQualityUCI.data) and skip the first line of data
• Create a method to remove rows that contain -200 or -2000 in column 3
• Change the plot x and y label
• Change t and k value 2 and 3 to 3 and 10
• Changed the number of hidden nodes from 3 to 4
• Changed the learning rate from 0.25 to 0.3