Misclassification of categorical exposure/treatment variables is prevalent in many fields of social and health sciences due to reporting errors. Many statistical approaches have been developed to address the bias of parameter estimates stemming from this issue. Typically, the information is obtained from external validation data containing the gold standard measurements that help to estimate the chances of misclassification. However, in many studies, gold standard measurements are not available, rather instrumental variables are accessible. In this work, we consider incorporating instrumental variables and other prior knowledge on the misclassification. We study the sufficient conditions for identifying model parameters and propose the use of a recently developed Bayesian method, ADVI, for estimation. We have conducted simulation studies to assess the operating characteristics of the method, and applied the method to US cancer mortality data sampled from the Surveillance Epidemiology and End Results (SEER) database.