This article mainly talks about some issues about SPSSPRO.
About: Following some online tutorials does not match the SPSSPRO results?
The reason for this is that many tutorials focus on the 0-1 binary classification problem, with clear positive and negative labels.
For example, if we're exploring the question of whether users churn, churn might be considered "positive" and retention "negative".
However, when we are faced with a basic binary classification problem such as "is it a cat or a dog", which does not determine the positive or negative for cats and dogs, or more complex multi-classification scenarios, things become less explicit, it is a normal classifier;
As a result, spsspro treats each level separately as if one class is positive and the rest are considered negative. In this way, you can get the above values for precision and recall for each category. What is ultimately given is the precision and recall after weighted averaging of the sample size for each category.
But now it can also support the calculation of confusion matrices and evaluation indicators for each classification level, we can go to the [Algorithm Sharing Library] to find the algorithm shared by others and can also see the [source**] of the calculation process, follow the steps below to !! againr/>
Step 1: Data in Machine Learning Classification.
Step 2: Upload the data, find the [Custom Algorithm] column in the data analysis module, click [Algorithm Library], find the algorithm [Multi-classification Machine Learning Evaluation], and click [Unlock Algorithm].
Step 3: After unlocking the algorithm, the algorithm will be generated under the custom algorithm list
Click Mode in the upper right corner to view the source of the algorithm.
Step 4: Drag in the variables of the corresponding variables to generate an output report.