# Task clustering

Clustering geographic features into operational tasks

# Rationale

There is often a requirement for grouping a set of mapped features (e.g. buildings, GPS locations etc.). For example, organizations may wish to cluster houses into operational groups as part of planning infrastructure projects. Similarly, those conducting field operations involving surveys or household visits may wish to group buildings into units which can then be assigned to field teams. While a number of off the shelf clustering algorithms exist, typically these have limitations which make them unsuitable for this task. For example, it may not be possible to limit the maximum numbers of features per cluster. Similarly, it may not be straightforward to incorporate natural divisions such as roads or rivers into the clustering process.

drawing Example of using Locational's clustering algorithm to cluster buildings in Eswatini

# Our approach

Locational's clustering algorithm allows users to create clusters of geographic features by setting the maximum distance between any two features. Users can also define the maximum number of features per cluster and provide a set of parcel features (lines) with which to parcel features before clustering. This ensures that no clusters intersect parcel line features. The algorithm itself uses a combination of distance-based clustering and k-means algorithms to iteratively solve the clustering problem given the user constraints.

# Implementations

This algorithm is being used by Reveal (opens new window) to cluster buildings into operational units for field teams delivering health interventions such as insecticide spraying for malaria and mass drug administration for neglected tropical diseases. The algorithm has also been used to identify clusters of wildlife tourism interest as part of work with WWF and the UN.

Think this sounds useful?

You can reach us at hello@locational.io to ask any questions, request additions or changes, or arrange a demo. We are actively developing these algorithms and would like to hear from you.