Metrics
This page summarizes metrics distributed with the SmartCGMS installation.
Simple metrics
SmartCGMS is distributed with a set of simple, well-known metrics:
- average error - an average value from all differences
- maximum error - a maximum value from all differences
- error at percentile - selects an error (difference) from a given percentile
- standard deviation (with Bessel correction) - standard deviation calculated from all differences
- variation (with Bessel correction) - variation calculated from all differences
- average error plus standard deviation (with Bessel correction) - calculates an average value and standard deviation from all differences and calculate the result as a sum of these two values
- RMSE - root mean square error from all differences
- area under ECDF - sums up an area under an ECDF
Advanced metrics
Apart from the simple metric set, the SmartCGMS distribution ships with some more advanced metrics.
Crosswalk
Based on: https://doi.org/10.1007/978-981-10-5122-7_221
Crosswalk is a time-ordered metric. We prefer to use this metric as it yields satisfactory results faster, than the common metric set.
Akaike Information Criterion
Based on: https://doi.org/10.1109/MSP.2004.1311138
Leal 2010
Based on: https://doi.org/10.1177/193229681000400221
This metric is an implementation of the "Best fit" equation from the above linked paper. However, as all metrics in SmartCGMS are considered distances from the ideal solution, the metric is inverted and the square root is dropped in favour of performance.