A Hybrid Approach for Outlier Detection in Weather Sensor Data

Abstract

IoT and big data technologies have embarked the modern data science. As nowadays lots of data have been generated from wireless sensors connected via a network. Detecting anomalous events in this large amount of data is the topic undergoing intense study among researchers. Most of the existing solutions for the detection of anomalous events in big data are based on machine learning models. The proposed technique is a hybrid approach to detect outliers in weather sensor data. The approach comprises of three phases. Initially, for handling big data efficiently, dimensionality reduction is performed in the first phase. In the second phase, the detection of anomalous events is done using multiple classifiers. Finally in the third phase, for final classification, the results of the different classifiers are combined. With the aid of the proposed approach, we can extract the meaningful information from a complex dataset. It can be perceived from the experimental results that the proposed approach outperforms the various state-of-the-art algorithms for outlier detection.