Categorical and numerical attributes occur frequently in cluster analysis tasks. To bridge the information gap between the heterogeneous categorical and numerical attributes in cluster analysis, the existing approaches usually adopt prior assumptions to distance definition and cluster distribution, which unavoidably introduce bias to the clustering process. To address this issue, we propose to analyze mixed data comprising both categorical and numerical attributes by forming minimal clusters through neighborhood set theory. As the minimal clusters are the smallest cluster units that can be obtained without relying on prior assumptions, unbiased cluster analysis can be facilitated accordingly. To avoid information loss, distance and density metrics that are unified on both numerical and categorical attributes are also proposed and utilized to merge the minimal clusters hierarchically. It turns out that our proposed approach is highly interpretable, and is capable of accurately and robustly clustering data sets composed of any combination of numerical and categorical attributes. Extensive experimental evaluations demonstrate its efficacy.