An Ensemble Learning Approach for Privacy–Quality–Efficiency Trade-Off in Data Analytics

Privacy is an issue of concern in the electronic era where data has become a primary source of investment for businesses and organizations. The value generated from data is put to use in a number of ways for economic benefit. Customer profiling is one such instance, where data collected is used for targeted marketing, personalized purchase recommendations and customized product deliveries. In such applications, the risk of individual sensitive information disclosure always prevails, affecting the privacy of individuals involved. Hence privacy preserving analysis demands suppressing or transforming data before it is published for analysis, thus curbing data leak. Subsequently, data quality degrades, and operative analytics is affected. With Big data, algorithms that offer a reasonable qualityprivacy trade off need enhancements in terms of efficiency and scalability. In this paper, the work proposed uses a privacy based composite classifier model to analyze the accuracy of classification. The diverse characteristics of algorithms in the composite classifier are found to balance the classification accuracy that is likely to get affected by privacy model. Further, the model’s performance with respect to execution time is then evaluated using the parallel computing framework Spark.