Density-based clustering has been widely studied by many researchers due to its popularity in generating arbitrary shaped clusters in data-stream. However, the majority of them are offline, sliding windowed or online-offline hybridization and cannot be applied to data-stream applications. This paper presents a fully online density-based clustering approach namely OCED (Online Clustering for Evolving Data-stream). The algorithm summarizes the data points in a sphere-shaped data structure called a micro-cluster. The initial radius of micro-cluster is learned and updated towards its optimal value based on the average data point density. A cluster graph is generated in an online manner using the micro-clusters to provide the arbitrary shaped cluster. The micro-clusters, which are irrelevant to recent data-stream content, are identified and removed to handle the evolving property. To the best of our knowledge, this is the first clustering approach that learns the radius instead of predefining. The approach is evaluated in a wide range of experiments. OCED proves its superiority over aligned approaches in terms of computational speed, cluster quality, and memory efficiency.