Cluster validation using an ensemble of supervised classifiers

Clustering Users Based on Hearing Aid Use: An Exploratory Analysis of Real-World Data

While the assessment of hearing aid use has traditionally relied on subjective self-reported measures, smartphone-connected hearing aids enable objective data logging from a large number of users. Objective data logging allows to overcome the inaccuracy of self-reported measures. Moreover, data logging enables assessing hearing aid use with a greater temporal resolution and longitudinally, making it possible to investigate hourly patterns of use and to account for the day-to-day variability. This study aims to explore patterns of hearing aid use throughout the day and assess whether clusters of users with similar use patterns can be identified. We did so by analyzing objective hearing aid use data logged from 15,905 real-world users over a 4-month period. Firstly, we investigated the daily amount of hearing aid use and its within-user and between-user variability. We found that users, on average, used the hearing aids for 10.01 h/day, exhibiting a substantial between-user (SD = 2.76 h) and within-user (SD = 3.88 h) variability. Secondly, we examined hearing aid use hourly patterns by clustering 453,612 logged days into typical days of hearing aid use. We identified three typical days of hearing aid use: full day (44% of days), afternoon (27%), and sporadic evening (26%) day of hearing aid use. Thirdly, we explored the usage patterns of the hearing aid users by clustering the users based on the proportion of time spent in each of the typical days of hearing aid use. We found three distinct user groups, each characterized by a predominant (i.e., experienced ~60% of the time) typical day of hearing aid use. Notably, the largest user group (49%) of users predominantly had full days of hearing aid use. Finally, we validated the user clustering by training a supervised classification ensemble to predict the cluster to which each user belonged. The high accuracy achieved by the supervised classifier ensemble (~86%) indicated valid user clustering and showed that such a classifier can be successfully used to group new hearing aid users in the future. This study provides a deeper insight into the adoption of hearing care treatments and paves the way for more personalized solutions.

Learning-Based Dissimilarity for Clustering Categorical Data

Comparing data objects is at the heart of machine learning. For continuous data, object dissimilarity is usually taken to be object distance; however, for categorical data, there is no universal agreement, for categories can be ordered in several different ways. Most existing category dissimilarity measures characterize the distance among the values an attribute may take using precisely the number of different values the attribute takes (the attribute space) and the frequency at which they occur. These kinds of measures overlook attribute interdependence, which may provide valuable information when capturing per-attribute object dissimilarity. In this paper, we introduce a novel object dissimilarity measure that we call Learning-Based Dissimilarity, for comparing categorical data. Our measure characterizes the distance between two categorical values of a given attribute in terms of how likely it is that such values are confused or not when all the dataset objects with the remaining attributes are used to predict them. To that end, we provide an algorithm that, given a target attribute, first learns a classification model in order to compute a confusion matrix for the attribute. Then, our method transforms the confusion matrix into a per-attribute dissimilarity measure. We have successfully tested our measure against 55 datasets gathered from the University of California, Irvine (UCI) Machine Learning Repository. Our results show that it surpasses, in terms of various performance indicators for data clustering, the most prominent distance relations put forward in the literature.

reval: A Python package to determine best clustering solutions with stability-based relative clustering validation

Determining the best partition for a dataset can be a challenging task because of the lack of a priori information within an unsupervised learning framework and the absence of a unique clustering validation approach to evaluate clustering solutions. Here we present reval: a Python package that leverages stability-based relative clustering validation methods to select best clustering solutions as the ones that replicate, via supervised learning, on unseen subsets of data. The implementation of relative validation methods can contribute to the theory of clustering by fostering new approaches for the investigation of clustering results in different situations and for different data distributions. This work aims at contributing to this effort by implementing a package that works with multiple clustering and classification algorithms, hence allowing both the automation of the labeling process and the assessment of the stability of different clustering mechanisms.