Efficient Model to Reduce False Positives using Outliers Detection in Big Data

Abstract

Emerging fields like Internet of Thing (IoT), sensor data, mobile computing, and social media are driving new forms and sources of data with distinct features. Big data is the term used to describe such type of data. Analytics of big data aims to use advanced analytic techniques on very large data sets that are collected from different sources in different formats. Mining anomalies from big data is a powerful mining task that is used mainly in critical systems. Applications work with big data require efficient outliers detection system. Outlier detection system in big data need to be efficiently designed to cope with its distinct features of volume, speed, complexity, and variety. The huge volume of outliers detected in big data is a barrier for outlier's diagnosis and analysis. Due to the cost of analysis each anomaly, outlier detector needs to be accurate as possible. Minimization of false positive alerts is a key feature that increases the accuracy of the detector system. This paper present a new propose model for reducing the false positive alerts using outliers detection. The proposed model uses cluster analysis by DBscan algorithm to highlight the outliers and then validates these outliers and reduces false positive alerts using Support Vector Machine (SVM) algorithm. The experimental study proves the efficiency of the proposed model with reported accuracy equals to 99%.