Machine Learning and AI Foundations: Classification Modeling

Classification methods are among the most important in modern data science. Learn classification strategies and algorithms for machining learning and AI.

One type of problem absolutely dominates machine learning and artificial intelligence: classification. Binary classification, the predominant method, sorts data into one of two categories: purchase or not, fraud or not, ill or not, etc. Machine learning and AI-based solutions need accurate, well-chosen algorithms in order to perform classification correctly. This course explains why predictive analytics projects are ultimately classification problems, and how data scientists can choose the right strategy (or strategies) for their projects. Instructor Keith McCormick draws on techniques from both traditional statistics and modern machine learning, revealing their strengths and weaknesses. Keith explains how to define your classification strategy, making it clear that the right choice is often a combination of approaches. Then, he demonstrates 11 different algorithms for building out your model, from discriminant analysis to logistic regression to artificial neural networks. Finally, learn how to overcome challenges such as dealing with missing data and performing data reduction.

Note: These tutorials are focused on the theory and practical application of binary classification algorithms. No software is required to follow along with the course.

Introduction

• Classification problems in machine learning
• What you should know
• Defining terms

1. The Big Picture: Defining Your Classification Strategy

• The importance of binary classification
• Binary vs. multinomial
• So-called “black box” techniques
• One task, many algorithms
• Statistics vs. machine learning
• Model assessment vs. business evaluation

2. How Do I Choose a "Winner"?

• Training and test partitions
• Lift Charts
• Gains tables
• Confusion matrix

3. Algorithms on Parade

• Overview
• Discriminant with three categories
• Discriminant with two categories
• Stepwise discriminant
• Logistic regression
• Stepwise logistic regression
• Decision Trees
• KNN
• Linear SVM
• Neural nets
• Bayesian networks
• Ensembles

4. Common Modeling Challenges

• Imbalanced target categories
• Interactions
• Missing data
• Bias-variance trade-off and overfitting
• Data reduction

Conclusion

• Next steps