Improve High Level Classification with a More Sensitive metric and Optimization approach for Complex Network Building

TL;DR

This paper proposes a novel method for high-level classification using class-specific complex networks, a new metric for network analysis, and an optimization approach that improves accuracy by up to 10%.

Contribution

It introduces class-specific network construction, a new network analysis metric, and an optimization strategy combining grid search and genetic algorithms.

Findings

Optimization improves classification results by up to 10%.

Class-specific networks outperform traditional all-sample networks.

The new metric effectively analyzes network structure.

Abstract

Complex Networks are a good approach to find internal relationships and represent the structure of classes in a dataset then they are used for High Level Classification. Previous works use K-Nearest Neighbors to build each Complex Network considering all the available samples. This paper introduces a different creation of Complex Networks, considering only sample which belongs to each class. And metric is used to analyze the structure of Complex Networks, besides an optimization approach to improve the performance is presented. Experiments are executed considering a cross validation process, the optimization approach is performed using grid search and Genetic Algorithm, this process can improve the results up to 10%.