To preserve patient confidentiality, there is a need to identify PHIs (Protected Health Information) from free text text clinical records, and such sensitive information must either be removed or replaced. Identification of the PHI's are normally performed manually on large sets of structured EHR databases, which is time-consuming, prohibitively expensive and error-prone. Hence, methods for automatic or semi-automatic identification of personal health information are of significant scientific and commercial interest. In this paper, we propose an innovative computational framework based on novel text mining and machine learning algorithms for automatic identification of PHIs from massive, unstructured free text clinical records, discharge summaries and other care documents. The experimental evaluation of the proposed algorithmic framework development, for several publicly available i2b2 challenge datasets from Informatics for Integrating Biology & the Bedside (i2b2) shared tasks, has shown promising outcomes.