No abstract

Most streaming decision models evolve continuously over time, run in resourceaware environments, and detect and react to changes in the environment generating data. One important issue, not yet convincingly addressed, is the design of experimental work to evaluate and compare decision models that evolve over time. This paper proposes a general framework for assessing predictive stream learning algorithms. We defend the use of prequential error with forgetting mechanisms to provide reliable error estimators. We prove that, in stationary data and for consistent learning algorithms, the holdout estimator, the prequential error and the prequential error estimated over a sliding window or using fading factors, all converge to the Bayes error. The use of prequential error with forgetting mechanisms reveals to be advantageous in assessing performance and in comparing stream learning algorithms. It is also worthwhile to use the proposed methods for hypothesis testing and for change detection. In a set of experiments in drift scenarios, we evaluate the ability of a standard change detection algorithm to detect change using three prequential error estimators. These experiments point out that the use of forgetting mechanisms (sliding windows or fading factors) are required for fast and efficient change detection. In comparison to sliding windows, fading factors are faster and memoryless, both important requirements for streaming applications. Overall, this paper is a contribution to a discussion on best practice for

No abstract

No abstract

This paper presents a system for induction of forest of functional trees from data streams able to detect concept drift. The Ultra Fast Forest of Trees (UFFT) is an incremental algorithm, which works online, processing each example in constant time, and performing a single scan over the training examples. It uses analytical techniques to choose the splitting criteria, and the information gain to estimate the merit of each possible splitting-test. For multi-class problems the algorithm builds a binary tree for each possible pair of classes, leading to a forest of trees. Decision nodes and leaves contain naive-Bayes classifiers playing different roles during the induction process. Naive-Bayes in leaves are used to classify test examples. Naive-Bayes in inner nodes play two different roles. They can be used as multivariate splitting-tests if chosen by the splitting criteria, and used to detect changes in the class-distribution of the examples that traverse the node. When a change in the class-distribution is detected, all the sub-tree rooted at that node will be pruned. The use of naive-Bayes classifiers at leaves to classify test examples, the use of splitting-tests based on the outcome of naiveBayes, and the use of naive-Bayes classifiers at decision nodes to detect changes in the distribution of the examples are directly obtained from the sufficient statistics required to compute the splitting criteria, without no additional computations. This aspect is a main advantage in the context of high-speed data streams. This methodology was tested with artificial and real-world data sets. The experimental results show a very good performance in comparison to a batch decision tree learner, and high capacity to detect drift in the distribution of the examples.

Gender, smoking habits, disease extent, need for corticosteroids, and hospitalization were all significantly associated with UC prognosis. These results may clarify the relative influences of these and other prognostic factors in the natural course of the disease and therefore help improve the management approach, thus improving the follow-up of patients.

Nowadays applications produce infinite streams of data distributed across wide sensor networks. In this work we study the problem of continuously maintain a cluster structure over the data points generated by the entire network. Usual techniques operate by forwarding and concentrating the entire data in a central server, processing it as a multivariate stream. In this paper, we propose DGClust, a new distributed algorithm which reduces both the dimensionality and the communication burdens, by allowing each local sensor to keep an online discretization of its data stream, which operates with constant update time and (almost) fixed space. Each new data point triggers a cell in this univariate grid, reflecting the current state of the data stream at the local site. Whenever a local site changes its state, it notifies the central server about the new state it is in. This way, at each point in time, the central site has the global multivariate state of the entire network. To avoid monitoring all possible states, which is exponential in the number of sensors, the central site keeps a small list of counters of the most frequent global states. Finally, a simple adaptive partitional clustering algorithm is applied to the frequent states central points in order to provide an anytime definition of the clusters centers. The approach is evaluated in the context of distributed sensor networks, focusing on three outcomes: loss to real centroids, communication prevention, and processing reduction. The experimental work on synthetic data supports our proposal, presenting robustness to a high number of sensors, and the application to real data from physiological sensors exposes the aforementioned advantages of the system.

No abstract