In lieu of activity recognition, the broader objective of a geriatric healthcare monitoring system is to identify cognitive impairments. The prerequisite of this identification is to model the behavioural pattern of the concerned person in terms of activity sequence. The behavioural pattern may change with the change in context parameters. Thus, a context-sensitive formal representation of behaviour is highly needed for apprehending the deviation within behavioural patterns. In this domain, most of the works are focused on activity discovery or activity recognition; minimal efforts are there on behavioural study. Moreover, the existing models handle simple transactional queries efficiently but for analytical queries, the situation is managed at run-time through the data retrieval from the lower-level recognition model. Instead of mining the entire dataset for each such query, an independent higher-level formal model for the representation of behaviour becomes effective. The caregiver can directly interact with the higher-level model for getting the output in less response time. In this paper, a graph-based behavioural analysis model is formally proposed on top of the underlying recognition model for achieving the above-said objective. The change in activity pattern for a specific context is carefully addressed within the proposed model. The model has been illustrated through a benchmark dataset and a rigorous experiment has been done on benchmark datasets ARUBA, TULUM, and KYOTO to measure the efficiency in terms of response time.