process.md

Machine Learning Process

Below are the steps that I have thought about when looking at the Myopia dataset. My overall goal is to go through a variety of machine learning methods in the process of determining a good quality model.

The most general description of my process is that I am looking for the best combination of Data Preparation, Learning Algorithm and Hyperparameters that make a representative model.

My Process:

1. Start with a Goal
2. Data Search
3. Exploratory Data Analysis
4. Analyze the Ask
5. Set Priorities and Limits
6. Select Intial Process (Model)
7. Re-evaluate the Goal and Value
8. Preprocess Data for Algorithm
   • Remove Unnecessary Features
   • Remove Outliers
   • Replace Missing Values
   • Balance Data
   • One-hot-encode Feature Classes
   • Label-encode Target Classes
   • Transform Data (for parametric algorithms)
   • Scale Data (distance or gradient descent or regularized algorithms)
9. Run Model
10. Evaluate Models
    • Classification Reports
    • ROC AUC or Precision-Recall AUC
11. Adjust Models
    • Changing Theshold Effects
    • Tuning Parameters
    • Remove Features
12. Select Best Model
13. Re-evaluate the Ask?