7.1 The optimization of data mining model should follow the principle of effective and moderate
- Effective principle: The conclusion of the model or whether the application effect satisfies the original business requirement
- Moderate principle: input/output cost performance
7.2 How to optimize the model effectively
7.2.1 Optimizing model from business thinking is the most important model optimization measure
- There is no more obvious and intuitive rules, indicators can replace complex modeling
- There are no obvious business logic (business assumptions) that have been overlooked in the pre-modeling phase
- Through early preliminary modeling and data familiarity, whether there are new discoveries, or even to subvert the previous business speculation or business intuition
- Whether the definition of the target variable is stable (sample validation at different points in time)
7.2.2 optimization from the technical thinking of modeling
7.2.3 optimization from the technical techniques of modeling 7.3 How to think about the limit of optimization 7.4 The main index system of Model effect evaluation (two yuan target variable)
7.4.1 Evaluation model accuracy and accuracy of series of indicators
- True Positive (TP): Refers to the number of observers that the model is predicted to be positive (1) and is indeed positive (1)
- True negative (TN): Refers to the number of observers that the model is predicted to be negative (0) and is indeed positive (0)
- False Positive (FP): Refers to the number of observers that the model predicts to be positive (1) and is actually negative (0)
- False negative (FN): Refers to the number of observers that the model predicts to be negative (0) and is actually positive (1)
The seventh chapter: Optimization and limitation of data mining modeling
