Top Page | Upper Page | Contents | About This Site | JAPANESE

Prediction by cluster analysis

Cluster Analysis is a method of dividing a sample into clusters (groups). Create clusters between objects with close coordinates.

After that, we asked questions such as "How was it divided?" and "Which group does this sample belong to?" It is common to analyze that.

However, clustering techniques can also be used for predictions by statistical models.

Each method is a step-by-step application of basic cluster analysis. In the figure below, "application" is represented by arrows.

How to use as unsupervised learning

Principal Component Analysis is unsupervised learning. There is a way to use a sample that was not used when the model was created as input data to see how the principal components are calculated. It's a similar usage.

• Analyzing Cluster Predictions: Analyzing : "Which Cluster Will It Fall In?"
• Analyzing Out-of-Cluster Predictions : "Can't I Fit into Any Cluster?"

The details of analyzing cluster predictions and analyzing out-of-cluster predictions are summarized in Analyzing Cluster Predictions.

How to use it as a pre-processing for supervised learning

Principal component analysis can be used as a preprocessing for explanatory variables when using supervised learning methods such as regression analysis. It's a similar usage.

• Label prediction : "Which label is more likely?"
• Numerical Prediction : "How many will it be?"

Label predictions and numerical predictions are summarized in Vector quantization label classification.

NEXT Analyzing Cluster Predictions