1. Variable Types

• Categorical Variables: Nominal and Ordinal
• Numerical Variables: Discrete and continuous
• Mixed variables: strings and numbers
• Datetime variables

2. Variable Characteristics

• Missing Data
• Cardinality
• Category Frequency
• Distributions
• Outliers
• Magnitude

3. Missing Data Imputation

• For Numerical Variables
  • Mean and Median Imputation
  • Arbitrary value imputation
  • End of Tail Imputation
• For Categorical Variables
  • Frequent category imputation
  • Adding a missing category
• Random Sample Imputation
• Adding a missing indicator
• Imputation with Scikit-learn
• Imputation with Feature-engine

4. Multivariate Imputation

• MICE
• KNN imputation

5. Categorical Variable Encoding

• Traditional Techniques
  • One hot encoding: simple and of frequent categories
  • Ordinal / Label encoding: arbitrary and ordered
  • Count / Frequency encoding
• Monotonic Relationship
  • Target mean encoding
  • Weight of evidence
  • Ordered label encoding
• Alternative Techniques
  • Binary encoding
  • Feature hashing
  • Probability Ratio
• For Rare Labels
  • One hot encoding of frequent categories
  • Grouping of rare categories
  • Rare Label encoding
  • Encoding with Scikit-learn
  • Encoding with category encoders

6. Variable Transformation

• Mathematical Transformations
  • Logarithic
  • Exponential / Power
  • Reciprocal
  • Box-Cox
  • Yeo-Johnson
• Discretisation
  • Unsupervised
    • Equal-width
    • Equal-frequency
    • K means
  • Supervised
    • Decision Tree
• Other
  • Transformation with Scikit-learn

7. Discretisation

• Arbitrary
• Equal-frequency discretisation
• Equal-width discretisation
• K-means discretisation
• Discretisation with trees
• Discretisation with Scikit-learn
• Discretisation with Feature-engine

8. Outliers

• Discretisation
• Capping / Censoring
• Trimming / Truncation

9. Feature Scaling

• Standardisation (common one)
• MinMaxScaling (common one)
• MaxAbsoluteScaling
• RobustScaling
• Scaling to absolute maxima
• Scaling to median & quantiles
• Scaling to unit norm

Models Effected by magnitude of feature

• Linear & Logistic Regression
• SVM
• KNN
• K-means Clustering
• LDA
• PCA
• Neural Networks

Models insensitive to feature magnitude - Tree Based Models

• Classification & Regression Trees
• Random Forest
• Gradient Boosted Trees

10. Mixed variables

• Creating new variables from strings and numbers

11. Datetime Variables

• Extracting day, month, week, semester, year ...etc
• Extracting hour, min, sec ...etc
• Capturing Elapsed time
  • Time between transactions
  • Age
• Working with timezones

12. Text

• Characters, Words, Unique words
• Lexical diversity
• Sentences, Paragraphs
• Bag of Words
• TFiDF

13. Transactions & Time Series

• Aggregate data
  • Number of payments in last 3, 6, 12 months
  • Time since last transaction
  • Total spending in last month

14. Feature Combination

• Ratio : total debt with income --> Debt to income ratio
• Sum : Debt in different credit cards --> total debt
• Subtraction : Income without expenses --> disposable income

15. Pipelines

• Classification Pipeline
• Regression Pipeline
• Pipeline with cross-validation