Fig. 4From: Pool detection from smart metering data with convolutional neural networksleft: Confusion matrix for the 5-nearest-neighbor classification with manual feature engineering from (Burkhart et al. 2018). The overall performance of the pool detection algorithm is 68.5% in terms of precision. The overall accuracy is 94.8%. right: Confusion matrix for the CNN classification: The overall performance is 71.9% in terms of precision. The overall accuracy is 95.5%Back to article page