As the mobile communication devices become popular, getting the location data of various objects is more convenient than before. Mobile groups that exhibit spatial and temporal proximities can be used for marketing, criminal detection, and ecological studies, just to name a few. Although nowadays the most advanced position equipments are capable of achieving a high accuracy with the measurement error less than 10 meters, they are still expensive. Positioning equipments using different technologies incur different amount of measurement errors ranging from 10 meters to a few hundred meters. In this thesis, we examine the impact of measurement errors on the accuracy of identified valid mobile groups and apply Kalman Filter and RTS smoothing as the one-way and two-way correction to correct the measurement data. In most settings, the corrected location data yield more accurate valid mobile groups. However, when the measurement error is small and users do not make abrupt change in their speed, mining mobile groups directly on the measurement data, however, yield better results.

As the advances of communication techniques, some services related to location information came into existence successively. On such application is on finding out the mobile groups that exhibit spatial and temporal proximities called mobile group mining. Although there exists positioning devices that are capable of achieving a high accuracy with low measurement error. Many consumer-grades, inexpensive positioning devices that incurred various extent of higher measurement error are much more popular. In addition, some natural factors such as temperature, humidity, and pressure may have influences on the precision of position measurement. Worse, moving objects may sometimes become untraceable voluntarily or involuntarily. In this thesis, we extend the previous work on mobile group mining and adopt Kalman filter to correct the noisy data and predict the missing data. Several methods based on Kalman filter that correct/predict either correction data or pair-wise distance data. These methods have been evaluated using synthetic data generated using IBM City Simulator. We identify the operating regions in which each method has the best performance.