Outlier Detection in Spatio-Temporal Data using Data Mining Technique

Abstract

Outlier detection is very important data mining method for various applications. The definition of outlier is that \an outlier is an observation that significantly deviates from other observations". Many outlier detection techniques have been developed for certain application domains, while others are developed as more generic. Outlier detection in spatio-temporal data are more challenging compared to identifying outliers in classical data. Spatio-temporal data that relate to both space and time. Spatio-temporal data mining refers to the process of discovering patterns and knowledge from spatio-temporal data. Spatio-temporal data are dense and highly correlated in nature. Here we are concentrating on outlier detection in spatio-temporal earth observation data. After ex- haustive literature survey we have identified various techniques and their advantages and disadvantages to detect outlier. Most of the techniques are focuses only on spatial aspect of the data but not the temporal aspect. These techniques were compared and based on the parameters like ability to detect arbitrary shaped cluster, time complexity, high dimensionality etc., we selected DBSCAN, ST-DBSCAN, ST-OUTLIER and SNN ap- proaches. Various input parameters were studied which is taken by these algorithms. Implementation of DBSCAN, ST-DBSCAN and our proposed approach ST-SNN is done using R Package (Open Source) and results are displayed using QGIS tool.