Search

In this era of big data, stream data classification which is one of typical data stream applications has become more and more significant and challengeable. In these applications, it is obvious that data classification is much more frequent than model training. The ratio of stream data to be classified is rapid and time-varying, so it is an important problem to classify the stream data efficiently with high throughput. In this paper, we first analyze and categorize the current data stream machine learning algorithms according to their data structures. Then, we propose stream data classification topology (SDC-Topology) on Storm. For the classification algorithms based on the matrix, we propose self-adaptive stream data classification framework (SASDC-Framework) for efficient stream data classification on Storm. In SASDC-Framework, all the data sets arriving at the same unit time are partitioned into subsets with the nearly best partition size and processed in parallel. To select the nearly best partition size for the stream data sets efficiently, we adopt bisection method strategy and inverse distance weighted strategy. Extreme learning machine, which is a fast and accurate machine learning method based on matrix calculating, is used to test the efficiency of our proposals. According to evaluation results, the throughputs based on SASDC-Framework are 8–35 times higher than those based on SDC-Topology and the best throughput is more than 40000 prediction requests per second in our environment. [ABSTRACT FROM AUTHOR]

Probabilistic threshold query (PTQ), which returns all the objects satisfying the query with probabilities higher than a threshold, is widely used in uncertain database. Most previous work focused on the efficiency of query process, but paid no attention to the setting of thresholds. However, setting the extreme thresholds may lead to empty result or too many results. It is difficult for a user to set a suitable threshold for a query. In this paper, we propose a new framework for PTQs based on threshold classification using ELM, where the probability threshold is replaced by the range of result number which is more intuitive and easier to choose. We first introduce the features selected for the two most important PTQs, which are nearest neighbor (NN) and reverse nearest neighbor (RNN) queries. Then a threshold classification algorithm (TCA) using ELM is proposed to set a suitable threshold for the query, where plurality voting method is applied. Further, the PTQ processing integrated with TCA are presented, and a dynamic classification strategy is proposed subsequently. Extensive experiments show that compared with the thresholds those the users input directly, the thresholds chosen by ELM classifiers are more suitable, which further improves the performance of PTQs algorithms. In addition, ELM outperforms SVM with regard to both the response time and classification accuracy. [ABSTRACT FROM AUTHOR]

The probabilistic group nearest neighbor(PGNN) query , which returns all the uncertain objects whose probabilities of being the group nearest neighbor (GNN) results exceed a user-specified threshold, is widely used in uncertain database. Most existing work for answering PGNN queries adopted a general framework which consist of three phases: spatial pruning, probabilistic pruning, refinement . In the probabilistic pruning phase, dividing the uncertain regions into many partitions to derive a tighter probabilities bounds is a common method. However, there is a tradeoff between the computational cost of probabilistic pruning phase and refinement phase controlled by the granularity of the partitions. In this paper, we study the problem of setting the optimal granularity of the partitions for uncertain objects, and propose a new framework for PGNN queries based on granularity classification using ELM such that the overall cost is minimized. In addition, to improve the accuracy of classification and make the classifier applicable to the dynamic environment, a plurality voting method and a dynamic classification strategy are proposed respectively. Extensive experiments shows that compared with the default granularities of the partitions, the granularities chosen by ELM classifiers are more proper, which further improves the performance of PGNN query algorithm. In addition, ELM outperforms SVM with regard to both the response time and classification accuracy. [ABSTRACT FROM AUTHOR]