Concept drift region identification via competence-based discrepancy distribution estimation

摘要: Real-world data analytics often involves cumulative data. While such contains valuable information, the pattern or concept underlying these may change over time and is known as drift. When learning under drift, it essential to know when, how where context has evolved. Most existing drift detection methods focus only on triggering a signal when detected, little research endeavored explain changes. To address this issue, we introduce kernel density estimation into competence-based method, invent discrepancy distribution identify specific regions in feature space occurred. Two experiments demonstrate that our proposed approach, estimation, can quantitatively highlight through space, produce results are very close preset regions.