Estimating Tukey depth using incremental quantile estimators

Measures of distance or how data points are positioned relative to each other fundamental in pattern recognition. The concept depth measures deep an arbitrary point is a dataset, and interesting this regard. However, while has received lot attention the statistical literature, its application within recognition still limited. To increase applicability recognition, we address well-known computational challenges associated with concept, by suggesting estimate using incremental quantile estimators. suggested algorithm can not only when dataset known advance, but also track for dynamically varying streams recursive updates. tracking ability was demonstrated based on real-life detecting changes human activity from real-time accelerometer observations. Given flexibility approach, it detect virtually any kind distributional patterns observations, thus outperforms detection approaches Mahalanobis distance.