project-parkinsons-disease-classification

This notebook uses the dataset from UCI Machine Learning Repository and classification models to predict whether a patient has Parkinson's Disease, with a focus on modele improvement on Logistic Regression.

Overview

I. Data Preparation

• Load and Oberserve Data
• Data Preprocessing
• Min-Max Normalization
• Problem Definition

II. Modeling

Each modeling section consists of an brief intro, the algorithm, discussion on related topics, and application on the dataset.

1. kNN
   • Non-parametric Models
   • Algorithm
2. Naive Bayes
   • Bayes Classifier
   • Algorithm
   • Generative Model vs. Discriminative Model
3. Logistic Regression
   • Sigmoid Function
   • Maximum Likelihood Estimation
   • Algorithm
   • (Also see Appendix A & B for related topics)
4. Support Vector Machine
   • Convex Sets and Convex Hulls
   • Algorithm
   • Soft-Margin SVM
5. Kernel SVM
   • Kernel
   • Mercer's theorem
   • RBF
   • Algorithm
6. Decision Tree

III. Model Improvement

• PCA
• Pipeline

IV. Model Selection

• ROC Curve
• Change Threshold
• Classification Report

V. Appendix

• Appendix A: Concepts for Logistic Regression

  • A1. Binary Classification
  • A2. Log Odds
  • A3. Linear Discriminant Analysis

• Appendix B: Linear Classifiers

  • B1. Definition
  • B2. Linear Separability
  • B3. Methods