Development and Implementation of Locally-Executed Data Mining on Smartphones

Data mining is the science of extracting information or ’knowledge’ from data. It is a task commonly executed on cloud computing resources, as well as personal computers and laptops. Nevertheless, what about smartphones? Despite the fact that these mobile devices now feature near-laptop levels of hardware and performance, locally-executed model-training using data mining methods on smartphones is exceedingly rare.
This thesis proposes a four-stage framework of analysis, research, implementation and expansion of data mining on mobile devices. First, it analyses the hardware capabilities of modern smartphones for locally-executing data mining algorithms through a series of empirical tests to learn how smartphones cope with this work. Second, it proposes two new data mining algorithms that focus on improving the model-training speed on mobile devices without losing classification accuracy. Third, it further proposes a practical mobile data mining solution that incorporates the above algorithms and operates on a broad range of smartphones without requiring cloud computing or internet resources. Finally, it proposes and implements a new novel protocol for multi-smartphone distributed data mining over private wireless networks to significantly improve model-training performance and combines it with a new mobile three-dimensional data exploration method using augment/mixed reality.
On-device data mining offers a number of advantages. It largely mitigates issues of data security and privacy, as no data is required to leave the device. It also ensures a self-contained, portable data mining solution that operates in any location and fits in the user’s pocket. In addition, the source code for our algorithms and data mining solutions are open-source and available as a basis of further research.