The problem of clustering time series has several applications in real-life contexts, especially in data science and data analytics pipelines. Existing time series clustering algorithms are ineffective for feature-rich real-world time series since they only compute the similarity of time series based on raw data or use a fixed set of features. In this paper, we develop a feature-based semisupervised clustering framework addressing the above issues for variable-length and heterogeneous time series. Specifically, we rely on a graph encoding of the time series that is obtained by considering a high number of significant extracted features. We then employ community detection and leverage a co-occurrence matrix in order to group together all the best clustering results. Our extensive experimental assessment shows the scalability and robustness of our approach along with its superiority against state of the art clustering algorithms on both real-world healthcare data and UCR benchmark data.