How does this work technically?
The Clustering process
The Clustering process starts as soon as you click on the Run / Train button in Trendskout. The system will execute various clustering algorithms on the input data, linked to the clustering analysis via drag & drop in the AI flow. Various combinations of algorithms and parameters are used herein, i.e. hypertuning.
2 criteria are crucial during the clustering process. First, the detected groups, clusters, must contain data points that are as close as possible to each other. Secondly, the number of clusters should remain limited. A clustering algorithm which detects groups of data that do not belong together, or that finds many small groups, means that further search is needed via hypertuning & Auto ML for better results. You can visualize the ratio between the number of clusters and the similarity in an “Elbow-curve”. v
Use of the Clustering output
A second use is where the clusters, and the underlying clustering model, are used to assign new data points to a cluster. This is comparable to classification. The difference with classification is that the training step is “Unsupervised”, and the labeling is based on automatically discovered groups or clusters.