Grant: Discovery Grants Program - Individual

With the recent advances in hardware and software, data streams have become ubiquitous as many sources produce data continuously and rapidly. Mining data streams is to extract knowledge structures or patterns in a non-stopping stream of data. Such patterns can be used for effectively monitoring, analyzing and predicting the changes in the application environment where the data stream is collected. Over the past few years, models and techniques have been proposed to extract patterns from data streams. However, many open issues still remain in data stream mining. The main objective of the proposed research program is to provide solutions to the open issues in data stream mining and develop a resource-adaptive framework for learning classification rules from data streams. We will work on the following issues. An efficient data structure will be designed that aggregates the data on the fly by making trade-offs between memory consumption and accuracy according to resource conditions. We will develop anytime rule learning and classification algorithms that can produce a best possible answer according to the real-time constraints. We will also systematically investigate the types of concept drifts and develop methods for detecting and handling different types of concepts drifts. In addition, we will design dynamic online discretization methods for use with rule learning from data streams. Mining data streams is still in its infancy state. The proposed research program will advance the field by providing solutions to its open challenges. Although we focus on classification rule learning in this research, the proposed data structure and algorithms can be adapted for use in other data stream mining tasks. With the growing proliferation of handheld devices, mobile phones, wearable computers and tablets, it becomes ever more important to develop online resource-adaptive data mining systems to assist people in decision making. The techniques developed in this research program will have a wide range of applications in areas such as networking, telecomminucations, scientific experiments, finance and medicine.