Local Correlation Integral Approach for Anomaly Detection Using Functional Data

The present work develops a methodology for the detection of outliers in functional data, taking into account both their shape and magnitude. Specifically, multivariate method anomaly called Local Correlation Integral (LOCI) has been extended adapted to be applied particular case using calculation distances Hilbert spaces. This validated with simulation study its application real data. taken scenarios data or curves different degrees dependence, as is usual cases continuously monitored versus time. results show that approach LOCI performs well inter-curve especially when are due magnitude curves. These supported by applying procedure meteorological database Alternative Energy Environment Group Ecuador, specifically humidity curves, presenting better performance than other competitive methods.