Detecting and Reacting to Changes in Sensing Units: The Active Classifier Case

Abstract

The ability to detect concept drift, i.e., a structural change in the acquired datastream, and react accordingly is a major achievement for intelligent sensing units. This ability allows the unit for actively tuning the application to maintain high performance, changing online the operational strategy, detecting and isolating possible occurring faults to name a few tasks. In the paper we consider a just-in-time strategy for adaptation: the sensing unit reacts exactly when needed, i.e., when concept drift is detected. Change detection tests (CDTs), designed to inspect structural changes in industrial and environmental data, are here coupled with adaptive k-NN and SVM classifiers, and suitably retrained when the change is detected. Computational complexity and memory requirements of the CDT and the classifier, precious limited resources in embedded sensing, are taken into account in the application design. We show that a hierarchical CDT coupled with an adaptive resource-aware classifier is a suitable tool for processing and classifying sequential streams of data.

Sliding Window-Based Fault Detection From High-Dimensional Data Streams

“…Depending on whether the raw data are labeled or not, knowledge-based techniques can be further classified into supervised and unsupervised ones [3]. The former uses plentiful positive (faulty) and negative (normal) data to learn the underlying data generating mechanisms for both classes, as has been done in [7], whereas the later learns the normal system behavior only from normal samples, and faults are detected as deviations from the learned normality, as has been done in [8]. Although supervised algorithms can provide favorable results in detecting and even isolating faults, faulty data for training purpose are generally insufficient and expensive to acquire in real-world applications.…”

“…3) Data streams can evolve as time progresses. This is also known as concept drifting [7], [24]. In the context of fault detection, the behavior of systems can vary over time-time-varying-due to many reasons, such as seasonal fluctuation, equipment aging, process drifting, and so forth.…”

Addressing Complexities of Machine Learning in Big Data: Principles, Trends and Challenges from Systematical Perspectives

“…The passive style of learning in non-stationary environments is to directly learn an evolving system that is just-in-time adaptive to the presence of changes in the distribution. These adaptive models include both single models [87][88][89] and ensemble models [90,91], but the latter tend to behave more stably and exhibit superior performance.…”

Intelligence for Embedded Systems

“…The output is t =T 2 , and [T 2 ,T ] is assumed to be generated by the process in the novel state, i.e., after concept drift cation, besides the CDT itself. Moreover, the H-CDT shows to be computationally lighter than CI-CUSUM [101] and in most applications involving embedded systems should be preferred. For this reason the code of the hierarchical CDT has been made freely available and can be downloaded from the link given in [102].…”