Benchmark and application of unsupervised classification approaches for univariate data

Maria El Abbassi,Michel Calame,Jan Overbeck,Herre S J Van Der Zant,Oliver Braun,Mickael L Perrin

doi:10.1038/s42005-021-00549-9

Abstract

Unsupervised machine learning, and in particular data clustering, is a powerful approach for the analysis of datasets and identification of characteristic features occurring throughout a dataset. It is gaining popularity across scientific disciplines and is particularly useful for applications without a priori knowledge of the data structure. Here, we introduce an approach for unsupervised data classification of any dataset consisting of a series of univariate measurements. It is therefore ideally suited for a wide range of measurement types. We apply it to the field of nanoelectronics and spectroscopy to identify meaningful structures in data sets. We also provide guidelines for the estimation of the optimum number of clusters. In addition, we have performed an extensive benchmark of novel and existing machine learning approaches and observe significant performance differences. Careful selection of the feature space construction method and clustering algorithms for a specific measurement type can therefore greatly improve classification accuracies.

Highlights

Unsupervised machine learning, and in particular data clustering, is a powerful approach for the analysis of datasets and identification of characteristic features occurring throughout a dataset
A schematic of the workflow for the unsupervised classification of univariate measurements is depicted in Fig. 1, starting from a dataset consisting of N univariate and discrete functions f(xi), i ∈ [1, N]
Each cluster validation indices (CVIs) provides a prediction for the number of clusters (NoC), after which the optimal NoC is estimated based on a histogram of the predictions obtained from all CVIs

Summary

Introduction

Unsupervised machine learning, and in particular data clustering, is a powerful approach for the analysis of datasets and identification of characteristic features occurring throughout a dataset It is gaining popularity across scientific disciplines and is useful for applications without a priori knowledge of the data structure. In all the above-mentioned studies, only a single feature space construction method and clustering algorithm were investigated, without a systematic benchmark of their accuracy against a large number of datasets of known classes and with varying partitions. This makes it difficult to compare the performance of one method to another. Few studies[25,31] provide guidelines for the estimation of the number of clusters (NoC), a critical step in data partitioning

Results

Discussion

Conclusion