Gemma Di Federico: Process Mining techniques as support for monitoring the worsening of dementia disease
Processes are everywhere, and we can model any process. Just think about the process of building a pen, the process of driving a car or the process of living a day. Deriving a model for each of them could be useful for different purposes, one of them is that we can continuously check if the activities are performed in the right and most probable way. For instance, in the healthcare field, patients effected by dementia usually fit into a very tight routine. A change in the habits can be correlated with an aggravation of the disease. The routine could be represented through a model, and the continuous observation of the process in the reality can lead to the identification of anomalous situations (with respect to the model). To do so, process mining techniques, typically used in the business domain, must be adapted and extended.
Process mining is an analytical discipline for discovering, monitoring, and improving processes as they actually are (not as you think they might be), by extracting knowledge from event logs readily available in today's information systems. The monitoring technique is called conformance checking, whose goal is to compare a log and a model to verify if the reality, as recorded in the log, conforms the model and vice versa. What is more, Internet of Things systems can be used to collect data. The integration between these two worlds, allows the development of more accurate, flexible and responsive business processes. Also moving towards other application scenarios, such as human behavior. In my PhD project, I would like to contribute into this transposition.
Human-related processes are unstructured and governed by uncertainty, differing from the typical business processes. Three main problems must be taken into consideration: the first refers to the fact that human being does not always perform actions in the same way, the second is that processes that arise from human activity are likely to experience concept drift, and all the analysis must be done online. Therefore, a modeling language and a conformance algorithm must be adapted. A solution could be the integration with probability functions which are a means capable to express dynamic behavior. The objective is to extend a (online) conformance checking algorithm such that it is able to evaluate the alignment between a stochastic model and a log, considering all the factors listed above.