Reweighted kNN classifiers and one-sided violence

Dnr:

SNIC 2017/1-300

Type:

SNAC Medium

Principal Investigator:

Måns Thulin

Affiliation:

Uppsala universitet

Start Date:

2017-07-01

End Date:

2018-01-01

Primary Classification:

10106: Sannolikhetsteori och statistik

Secondary Classification:

50901: Tvärvetenskapliga studier (t.ex Freds- och konfliktforskning och studier om ett hållbart samhälle)

Webpage:

http://www.mansthulin.se/research/

Allocation

Abstract

Many classification methods (including kNN classifiers) are not invariant under rescaling, meaning that rescaling the data can change the results. In this project, we investigate and evaluate novel methods for finding rescalings that improve the performance of the classifiers. The resulting methods will be used to predict one-sided violence events, using a database from the Department of Peace and Conflict Research at Uppsala University.