The target of the process analysis for abnormal condition is , for example, "Why does the distribution happen?" and "Why did the wrong product be made?". Those are found in the process analysis for normal condition .
Before starting analysis of X, we should grasp what happened. "What happened" is the information of Y. The first step is studying Y, the second step is analyzing X.
For example, if the problem is the unknown matter in the products, before looking for the machine of the cause, I analyze the ingredients and the shape of the matter. This information will be powerful hint.
Before the step of the process analysis, we need to collect information about the fact. In many cases, we need much time. The success of data analysis depends on these steps.
It is important to look at "Data" itself because the change of the style of the data may be the cause of abnormality. The basic of Data Science is to look at data itself.
Graphs are useful to look at many numbers.
Using mathematical model is less important. Without looking at data, mathematical approaches do not work well because actual data has complicated background.
Mathematical models of regression analysis and discriminant analysis uses formulation and normal distribution . This approach does not match for raw data in many cases.
We need stratified sampling or data cleansing to use this approach.
Data cleansing is often used to delete abnormal data. But for the analysis of abnormality, using abnormal data may work well. Deleting abnormal data is used to compare mathematical model of normal condition and abnormality.
The idea of Decision tree is the key to analyze. It divides samples.
MT method can express abnormality as quantity data.
Statistical Way of Making Hypothesis
Outlier and Missing Value
NEXT Cause analysis of temperature and humidityTweet