Clustering of Bi-Dimensional and Heterogeneous Times Series: Application to Social Sciences Data

We present an application of bi-dimensional and heterogeneous time series clustering in order to resolve a Social Sciences study issue. The dataset is the result of a survey involving more than eight thousand handicapped persons. Sociologists need to know if there are in this dataset some homogeneous classes of people according to two attributes: the idea that handicapped people have about the quality of their life and their couple status (i.e. if they have a partner or not). These two attributes are time series so we had to adapt the k-Means clustering algorithm in order to be efficient with this kind of data. For this purpose, we use the Longest Common Subsequence time series distance for its efficiency to manage time stretching and we extend it to the bidimensional and heterogeneous case. The results of our unsupervised process give some pertinent and surprising clusters that can be easily analyzed by sociologists.Présentation d'une application d'un "bi-dimensional and heterogeneous time series clustering" pour résoudre un problème en sciences sociales. Les données concernent plus de huit mille personnes en situation de handicap. Le problème est de savoir s'il existe de groupes homogènes vis-à-vis de la qualité de vie ressentie et de la vie de couple déclarée. A ces deux séries temporelles, un algorithme de k-Means clustering a été adapté. Nous avons utilisé the Longest Common Subsequence time series distance et nous l'avons étendue au cas bi-dimensionnel et hétérogène. Le résultat a été pertinent et surprenant, utile à l'analyse sociologique