Infectious disease models are important tools for informing public health policy and gaining a better understanding of disease transmission dynamics. For them to accomplish these goals effectively it is important that models accurately represent the mechanisms that drive disease spread. During many epidemics, changing human behaviour is a driving force behind shifts in transmission patterns. In times of perceived higher risk people will often adopt protective behaviour changes to attempt to mitigate the impact of the disease on themselves or others in their community. These behaviour changes are typically complex, not well understood, and difficult to quantify. For these reasons behaviour change is challenging to incorporate into an infectious disease model. However, ignoring its effect can lead to inaccurate estimates of epidemic metrics and misleading forecasts of future trajectories. It is therefore important to develop methods that allow for the effect of behaviour changes to be adequately captured. We propose a flexible framework for incorporating epidemic-driven behaviour change into an individual-level model (ILM), which models individual probability of infection over time as a function of individual-level covariates. Focusing on spatially-based ILMs, we consider four potential so-called “alarm functions” for relating prevalence to population protective behaviour changes. We explore the impact of allowing behaviour changes to affect different model components, and of misspecifying the form of the alarm function or model structure. Methods are applied to a data set of the 2001 foot and mouth disease epidemic amongst livestock in the United Kingdom. Next we outline a framework to allow dynamic time-dependent adherence to behaviour changes in a population-averaged framework. Four adherence functions are proposed and investigated through simulation studies. These “dynamic behaviour change models” are then used to analyse four waves of COVID-19 in Calgary, Alberta. Finally, we conduct a data analysis of two waves of COVID-19 in four locations throughout Canada and the United States. We compare models that have either an epidemic history- or behaviour proxy covariate-driven behaviour change effect, or a behaviour change effect that is informed by both epidemic history and an external covariate.