Accumulating regional density dissimilarity for concept drift detection in data streams

TLDR: The overall results show that NN-DVI has better performance in terms of addressing problems related to concept drift-detection, including both synthetic and real-world datasets.

About: This article is published in Pattern Recognition.The article was published on 2018-04-01 and is currently open access. It has received 82 citations till now. The article focuses on the topics: Concept drift & Data stream mining.

Figures

FIGURE 4: Using a k-nearest neighbor data presenting model to replace the hyperspheres model, the arrow in (b) indicates the relationship of k-nearest neighbors. For example, d1 → d4 indicates that d4 is a knn member of d1 while d1 is not a knn member of d4. As long as there is a connection between two data instances, we consider that they are neighbors. Then the datasets can be presented as a set of connections, as shown in (c). If we consider each connection as one component of a data instance, namely a slice of instance, then the data instance d1 is a composite of one slice of itself, one slice from d2 and one slice from d4, as shown in (d). Then, (e) illustrates the difference between conventional partitioning methods and the NNPS

TABLE 3: Drift detection results on P (4) stream

TABLE 4: Drift detection results on HD C(4) streams

TABLE 5: Average Drift Detection Results

FIGURE 8: The dnnps between normal distributed data (400 instances each time step) with varying µ. For each odd timestep, we compared two i.i.d. sample sets, while for each even timestep we compared two sample sets with drifted mean (µ+ 0.06). All the statistics were computed based on the average of 100 runs.

FIGURE 10: The test statistics of two-sample K-S test between normal distributed data with varying σ. For each odd timestep, we compared two i.i.d. sample sets, while for each even timestep we compared two sample sets with drifted standard deviation (σ + 0.02). All the statistics were computed based on the average of 100 runs.