Overview

Classic classification by local modeling is a two-step approach for modeling:

1. An unsupervised clustering algorithm is run to find regions in the dataset;
2. For each region, a model is built with the respective data subset.

For inference the procedure is similar:

1. A similarity metric is used to determine the new data point region, e.g. euclidian distance from regions prototypes;
2. The model from that specific region is used to predict the class of the new data point.

To install dependencies run pip install -r requirements.txt on the main directory.

Built with:

• Pandas
• Numpy
• Sklearn
• Plotly

You can see the notebook here.