During the evolution of a software system, it is inevitable that changes accumulatively lead to architecture erosion and architectural debt. A plethora of metric-based methods have been developed to identify architectural problems, which violate the design principle and degrades the software maintainability. However, as argued by practitioners, those metrics with the architectural problems detected by them are lack of the comprehensibility, interpretability, and indicative(lity) when applied to the industry. To address this issue, we propose dbMIT to identify potential code changes that make architecture decay. Our dbMIT first integrates popular implementation-level metrics such as the CK suite through syntactic dependencies and evolutionary dependencies. Then dbMIT constructs a forest structure as a knowledge base to relate the metrics together, where the relations go through the multiple-granularity metrics. Via pre-defined rules using the knowledge base, our dbMIT identifies code changes potentially causing the architecture decay. Due to usage of the forest structure of code-level metrics, it is easy for developers to understand the measurements, to explain why the detected code changes are potential contributors for the decay, and to indicate how to resolve them. Based on dbMIT, we also contribute a web-based tool to measure, detect and visualize the software architecture with potential causes of the maintainability difficulty. Our experiments on the collected open-source projects demonstrate the effectiveness of dbMIT in comparison with existing methods.