Reliability estimation (Orange.evaluation.reliability)

Reliability assessment statistically predicts reliability of single predictions. Most of implemented algorithms for regression are taken from Comparison of approaches for estimating reliability of individual regression predictions, Zoran Bosnić, 2008. Implementations for classification follow descriptions in Evaluating Reliability of Single Classifications of Neural Networks, Darko Pevec, 2011.

The following example shows basic usage of reliability estimation methods:

The important points of this example are:

It is also possible to do reliability estimation on whole data table, not only on single instance. Next example demonstrates usage of a cross-validation technique for reliability estimation. Reliability estimations for first 10 instances get printed:

Reliability Methods

For regression, all the described measures can be used, except for the $$O_{ref}$$ . Classification domains are supported by the following methods: BAGV, LCV, CNK and DENS, $$O_{ref}$$ .

Reliability estimation results

There is a dictionary named :obj:METHOD_NAME that maps reliability estimation method IDs (ints) to method names (strings).

In this module, there are also two constants for distinguishing signed and absolute reliability estimation measures:

SIGNED = 0
ABSOLUTE = 1


Example of usage

This script prints out Pearson's R coefficient between reliability estimates and actual prediction errors, and a corresponding p-value, for each of the reliability estimation measures used by default.

Estimate               r       p
SAvar absolute        -0.077   0.454
SAbias signed         -0.165   0.105
SAbias absolute       -0.099   0.333
BAGV absolute          0.104   0.309
CNK signed             0.233   0.021
CNK absolute           0.057   0.579
LCV absolute           0.069   0.504
BVCK_absolute          0.092   0.368
Mahalanobis absolute   0.091   0.375


References

Bosnić, Z., Kononenko, I. (2007) Estimation of individual prediction reliability using local sensitivity analysis. Applied Intelligence 29(3), pp. 187-203.

Bosnić, Z., Kononenko, I. (2008) Comparison of approaches for estimating reliability of individual regression predictions. Data & Knowledge Engineering 67(3), pp. 504-516.

Bosnić, Z., Kononenko, I. (2010) Automatic selection of reliability estimates for individual regression predictions. The Knowledge Engineering Review 25(1), pp. 27-47.

Pevec, D., Štrumbelj, E., Kononenko, I. (2011) Evaluating Reliability of Single Classifications of Neural Networks. Adaptive and Natural Computing Algorithms, 2011, pp. 22-30.