Relational Forecast Limiter Algorithm for ICD based EMRs

Abstract

Forecasting individual privacy based on acquired knowledge across associated or related diseases is often a concern during data publishing. EMR (Electronic Medical Record) of an individual have more than one associated diseases which are potential knowledge nodes for analyst to exploit privacy. RFL (Relational Forecast Limiter) algorithm aims in reducing the forecast or prediction level by introducing generalization, suppression and noise addition techniques based on relational forecast detection and relational forecast height of diseases classified in ICD (International Classification of Diseases) table. These techniques delimit the forecasting capability to bring privacy under control. Generalization and suppression techniques are applied on sensitive attributes while noise addition is applied on quasi identifiers. Generalization is realized at lower twig and branch level, while suppression and noise addition are realized at bough level. Primary objective of the algorithm focuses on sharing minimum privacy data enabling the data analyst to extract maximum useful information. Accuracy is retained to ensure data analysis yields useful information for social causes. Experimental results on privacy and accuracy loss demonstrates algorithm efficiency.