Evaluation of classification of Building Time Series with Matrix Profile

Supervisors: Julien Nembrini, Denis Lalanne

Student: Nicolas Bovet

Project status: Finished

Year: 2022

A recent method called Matrix Profile has gained some recognition in data classification with the advantage of being very fast and deterministic. The objective of this work is to explore the use of the Matrix Profile on building data using the MPF library for python which proposes a hierachical clustering with Matrix Profile. We compare the impact of different resampling strategy on the classification by means of three clustering methods : one classical method with kmeans and two that use Matrix Profile: one of the MPF library for hierarchical clustering, and one using graph theory.

Document: report.pdf