Analyzing and mining multidimensional data streams

Abstract

Data streams are usually characterized by large amounts of data generated in continuous, potentially infinite, synchronous or asynchronous processes, in applications such as: meteorological systems, industrial processes, vehicle monitoring systems, financial transactions, sensor networks, among others. Moreover, the behavior of the data tends to change significantly over time, defining evolving data streams. These changes may mean temporary events (such as anomalies or extreme events) or relevant changes in the process of generating the stream, resulting in changes in the distribution of data (climate changes, for instance). This project aims to develop techniques for analysis and mining of evolving, multidimensional data streams, focusing on knowledge discovery from agro-meteorological data, such as: real climate measures collected from meteorological stations, climate measures generated by climate models and remote sensing data related to agricultural monitoring. The initial approach is based on concepts of the Fractal Theory, applied to analyze temporal behavior. The application in agro-meteorological data aims to identify extreme weather events, climate changes and the impact of these events in coffee and sugarcane growing areas in Southeastern Brazil. Therefore, this work can potentially result in contributions to the data mining area as well as support research on Agrometeorology. (AU)