What is it about?
In many sectors, classifying an observation accurately with the desired balance between precision and sensitivity (or recall) is preferred. Classification based on a set of features of neighboring observations is a common approach simple for interpretation. This paper proposes to optimize the balanced metric between precision and sensitivity, known as Fbeta score, by the k-nearest neighbor algorithm and accounts for the choice of features of neighbors.
Featured Image
Photo by Myles Bloomfield on Unsplash
Why is it important?
The proposed optimization approach for the k-nearest neighbor algorithm is tested in six public datasets of binary classification problems and compared with an ensemble approach. When a decision-maker considers sensitivity (or recall) equally or more important than precision, the proposed approach, using a subset of features of neighboring observations, has shown better performances (of Fbeta scores) in most of the six public datasets than the ensemble approach.
Perspectives
Existing algorithms mainly adopt a one-way approach to select a full/partial set of features from a given set of observations as inputs and transform them into outputs targeting at the observed classes. An optimization framework can incorporate both inputs, outputs into a two-way model and enable one or multiple performance metrics to be optimized in the objective function or constraints.
Ka Yuk Carrie LIN
City University of Hong Kong
Read the Original
This page is a summary of: Optimizing hyperparameters in the k-NN classifier to maximize Fβ score, January 2025, American Institute of Physics,
DOI: 10.1063/5.0286046.
You can read the full text:
Contributors
The following have contributed to this page
