The higher level: The distinction between having or not a reference variable to be explained
   - No response variable: non-supervised methods
   - With response varialbe: supervised methods

The second level:
   For methods oriented to description,
    - Descriptive methods: describing relationship between objects (raws of data matrix)
    - Associative methods: describing relationship between variables (columns of data matrix)
   For methods oriented to prediction,
    - Descriminant methods: predicting qualitative variables
    - Predictive methods: predicting quantitative response variables

Discriminatn models have one more level of subdivision:
   - Rule-based reasoning methods: methods providing explicit knowledge model, which can be expressed by formal rules
   - Case-based reasoning methods: the predictive model is implicit in historical data
   - Bayesian learning methods: a mixture between explicit knowledge model and iterative refinement based on future data