Analyzing Adaptive and Non-Adaptive Online Learners on Imbalanced Evolving Streams

Recently, the combined problem of online class imbalance and concept drift (OCI-CD) has received much interest. The effect this on state-of-the-art adaptive non-adaptive learners is not widely investigated. This work explores impact parameters such as current ratio, length stream, type drift, levels state (static or dynamic) learners. experimental results demonstrate that each parameter considered for study a notable learner performance: (a) minority performance decreases with increase degree imbalance, (b) are prone to both drifts than learners, (c) only susceptible drifts, (d) dynamic more adverse static (e) large scale support vector machine yields stable all study. Further, directions develop new approaches also presented based these findings.