10.3 Lec2 CRISP-framework
Significant point of this Lec (SP):
• The difference between explanatory modeling and predictive modeling; pg4-19
1. Different goal
2. Different evaluation
3. Different modeling paths
• [NB]Data preprocessing:
1. its motivation or reason: why we should do it; pg21
2. what should we do and how (sampling, encoding, missing values,
outliers, normalizing…) ;pg22-46
• The difference between types of variables
Highlight:
• Slides: pg4-19, pg 21, pg 22-46
• Books:
Info:
11.15 visit AXA, register first; ?? TBC project upload; 12.12 ceremony competition
1 The difference between Exp vs Pre modeling
Goals!DEFINITION"
• Explanatory modeling: Theory-based, statistical testing of causal hypothesis
• Predictive modeling: Data science methods to make predictions
Evaluation
• Explanatory modeling: Strength of relationship in statistical model
1
, • Predictive modeling: Ability to accurately predict new observations
Modeling path: (17:00--)
• Data collection\ data preparation\data partitioning (important! Next week)
1. Data collection, similar
2. Data preparation, facing data missing—explanatory modeling can throw it
away; but for predictive modeling, it’ll be a problem.
3. Data partitioning, not important for explanatory but super important for
predictive modeling. (more info:
https://www.cockroachlabs.com/blog/what-is-data-partitioning-and-
how-to-do-it-right/)
• About the choice of variables:
1. for explanatory modeling, operationalized variables serve as practical
instruments for investigating the underlying conceptual constructs and
the relationships between them. For example, a questionnaire designed
to assess a person's level of depression (the construct) by asking about
their feelings and behaviors is a practical instrument. The term is used
often in the social sciences because scientists in that field have to spend
so much time creating and validating their constructs of interest, just to
be able to measure for them.)
2. for predictive modeling, the variables can be way broad, hundreds to
thousands, of course those should be available at first.
Notable words:
collaborative filtering models
2
, Definition: Collaborative filtering filters information by using the
interactions and data collected by the system from other users. It’s
based on the idea that people who agreed in their evaluation of certain
items are likely to agree again in the future. The algorithm supports
recommended system, for example, Taobao , amazon, netflix #$%
&'
Other differences:
• Explaining does not necessarily lead to predictions: variables nor present.
• Multicollinearity is a problem in explanatory model but not usually in predictive
modeling. Multicollinearity will not affect the ability of the model to predict. (A
websites clearify this: https://hackernoon.com/multicollinearity-and-its-
importance-in-machine-learning)
• Method:
1. explanatory—interpretable statistical method;
2. predictive—accurate machine learning method.
• Validation:
1. Model fit and R2. R2 is a measure of the goodness of fit of a model.[11] In
regression, the R2 coefficient of determination is a statistical measure of how
well the regression predictions approximate the real data points. An R2 of 1
indicates that the regression predictions perfectly fit the data.
2. Generalisation and accuracy.
• Y=f(X), to explain, test a given f; to predict, find f.
3
, 2 Data preprocessing
• Motivation and reason: dirty and noisy data, inconsistent data, incomplete data…
1-Data preprocessing. Sampling:
1. Definition: Select a suitable or representative sample to determine the
parameters and characteristics of the whole population.
2. Reason: economic, time, large and partly accessible population, computation
power.
3. How to do sampling and things to avoid: stratified sampling ()*+,
timing!many data vs recent/relevant data",avoid seasonality
effects(sales during summer and winter).
2-Data preprocessing. Encoding:
1. Encoding is the process of converting categorical data into a numerical
format that machine learning algorithms can understand.
2. Encoding vs Normalizing
4