Machine Learning with R

Part 1: Overview of Data Science

1. Data Science as a quantitative discipline
   • How to define Data Science scopes
   • The many faces of Data Science: Data Mining, Data Analysis, Data Analytics, Machine Learning, Predictive Modeling, Statistical Learning, Mathematical Modeling. What are these all about?
   • Data Mining as a data exploration process
   • Machine Learning: supervised vs. unsupervised
   • Machine Learning vs. Predictive Analytics
   • Big Data Analytics: what is it and why it’s important
2. Overview of a Data Mining process cycle
   • Understanding business needs and identifying new business opportunities
   • Formulating a business problem and associated requirements
   • Defining key quantitative metrics to measure success and evaluating business benefits
   • Translating business requirements into technical requirements and documentation
   • Formulating data models based on business and technical requirements
   • Identifying a set of quantitative models based on technical requirements and metrics of success
   • Running the models and evaluating results
   • Selecting the best model
   • Deploying the model

Part 2: The Data Foundation

1. Data sources
2. Types of data
   • Structured vs. unstructured data
   • Static data vs. real-time data
   • Types of data attributes: numerical vs. categorical
   • Role of time factor and time trends in data analysis
3. Working with missing values
   • Main causes of missing data
   • Understanding the importance of missing information
   • Types of missing information
   • Restoring missing values
   • Imputing missing values and selecting imputation techniques
   • Understanding and evaluating potential consequences of manipulating records with missing values
4. Working with outliers
   • Defining quantitative criteria for outlier detection in 1D cases
   • Understanding role of outliers in model building
   • Deciding on outlier removal
   • Defining outlier detection metrics in multi-dimensional space
5. Working with duplicate records
   • Defining duplicates
   • Understanding sources of duplicates
   • Deciding on duplicate removal

Part 3: Sampling and Hypothesis Testing

1. Why sampling may be important for Machine Learning
2. Sampling techniques and sample bias
3. Statistical hypothesis
4. Z-score, t-score and F statistic
5. P-values
6. Implementation of hypothesis testing for model evaluation analysis

Part 4: Machine Learning Fundamentals

1. What is Machine Learning?
2. Supervised vs. unsupervised learning
3. Overview of supervised Machine Learning
   • Regression models
   • Classification models
4. Overview of unsupervised Machine Learning
   • Clustering methods
   • Principal component analysis and dimension reduction
   • Association rules
5. Overview of major steps in building and testing quantitative models
   • Criteria for model selection
   • How to prepare a training set
   • Criteria for selecting model attributes/predictors
   • Working with collinear variables
   • Addressing imbalance problem
   • Dealing with over-fitting; bias-variance tradeoff
   • Validation and cross-validation

Part 5: Building a Linear Regression Model with R.

1. Univariate regression vs. multiple regression
2. Mathematical foundation of linear regression overview: least square method vs. maximum likelihood method
3. Model assumptions
4. Working with continuous attributes
5. Dealing with collinear variable
6. Model subset selection:
   • Forward stepwise selection
   • Backward selection
   • Shrinkage methods: ridge regression and Lasso
   • Dimension reduction
   • Information criteria
7. Automating model selection procedure
8. Model parameter evaluation, R squared vs. adjusted R squared
9. Validating the model
10. Working with categorical variables
11. Considering input variable interactions

Part 6: Example of building a Classification Model with R

1. Dealing with imbalanced training sets
2. Understanding confusion matrix
3. Evaluating binary classifiers using ROC / AUC

Part 7: Example of Cluster Analysis with R

1. Overview of cluster analysis mathematical foundation
2. K-means clustering method
   • Algorithm overview
   • Convergence criteria
   • How to determine the number of clusters

Part 8: Dimension Reduction techniques with R

1. What is dimension reduction?
2. The practical goals of dimension reduction implementation
3. Principal component analysis vs. singular value decomposition
4. How many components to choose

Part 9: Class Conclusion

1. What was not covered in the class
2. Big Data Analytics – the future of machine learning: main tools and concepts