Big data is data set that cannot reasonably be handled by our traditional database or tools due to the large volume, velocity and variety. In this vast and complex data set outlier detection is very crucial aspect. A failure to detect outliers or their ineffective handling can have serious ramifications on the strength of the inferences drained from the exercise. This dissertation proposes a reliable and high throughput outlier detection technique which attempts to detect projected outlier in high dimensional data stream. Optimize ORION algorithm employs an innovative window based time model in capturing dynamic statistics from stream data, and novel data structure containing a set of top sparse subspaces to detect projected outlier effectively in low process time and less memory storage. This algorithm able to identifies a data point as an outlier if it resides in low density region. Optimized Orion Increase throughput 20% and enhance stability 25%.
