In a federated learning system, the clients, e.g. mobile devices and organization participants, usually have different personal preferences or behavior patterns, namely Non-IID data problems across clients. Clustered federated learning is to group users into different clusters that the clients in the same group will share the same or similar behavior patterns that are to satisfy the IID data assumption for most traditional machine learning algorithms. Most of the existing clustering methods in FL treat every client equally that ignores the different importance contributions among clients. This paper proposes a novel weighted client-based clustered FL algorithm to leverage the client's group and each client in a unified optimization framework. Moreover, the paper proposes convergence analysis to the proposed clustered FL method. The experimental analysis has demonstrated the effectiveness of the proposed method.