Fateme Farhadi - University of Zabol
Saeed Arbabi - University of Zabol

There is an emerging focus on real-time data stream analysis on mobile and ubiquitous devices. A wide range of applications that process data streams are targeted to run on mobile handheld devices with limited computational capabilities. The streams of data required to be analyzed are turn out at the data sources. Data sources include contextual data about users and how they use their phones (e.g. user location, application and battery usage, online activity, call and SMS behavior, charging behavior) and sensors which perform monitoring/measurements and transmit this data continuously such as environment sensors that measure physical phenomena (e.g. temperature, pressure, soil-humidity), or biosensors that measure physiological phenomena (e.g. heart-rate ECG, movement levels etc.).Given the small size, portability, and mobility of users, mobile and ubiquitous devices are limited in terms of battery power, computation, communication, and visualization facilities. Key restrictions of such environment include bandwidth, CPU cycles, memory, storage, visualization, mobility, and connectivity. Address these challenges, a more important issue is the quality of the approximate mining results. More accurate results usually need more memory and computational resource. Although advancements in micro- and nano-chipset technologies have empowered mobile devices, these resource-constraints need to be considered in the development of data mining algorithms.In this paper, we discuss challenges that are in previous stream mining algorithms and present a hybrid algorithm to mine itemsets from a transaction stream in personal devices. This algorithm addresses these challenges and leverage by two pass in order to efficiently use memory.

Data Stream, Mobile Mining, Data Analysis, Stream Mining