Accuracy Paradox

From GM-RKB
Jump to navigation Jump to search

The Accuracy Paradox states that predictive models with a (lower) level of accuracy may have greater predictive power than those with higher level of accuracy.



References

2016

  • (Wikipedia, 2016) ⇒ http://external Retrieved 2016-07-10
    • The accuracy paradox for predictive analytics states that predictive models with a given level of accuracy may have greater predictive power than models with higher accuracy. It may be better to avoid the accuracy metric in favor of other metrics such as precision and recall.

      Accuracy is often the starting point for analyzing the quality of a predictive model, as well as an obvious criterion for prediction. Accuracy measures the ratio of correct predictions to the total number of cases evaluated. It may seem obvious that the ratio of correct predictions to cases should be a key metric. A predictive model may have high accuracy, but be useless.

      In an example predictive model for an insurance fraud application, all cases that are predicted as high-risk by the model will be investigated. To evaluate the performance of the model, the insurance company has created a sample data set of 10,000 claims. All 10,000 cases in the validation sample have been carefully checked and it is known which cases are fraudulent. To analyze the quality of the model, the insurance uses the table of confusion. The definition of accuracy, the table of confusion for model M1Fraud, and the calculation of accuracy for model M1Fraud is shown below.

2014

  • (Valverde-Albacete and Peláez-Moreno,2014) ⇒ Valverde-Albacete, F. J., & Peláez-Moreno, C. (2014). 100% classification accuracy considered harmful: The normalized information transfer factor explains the accuracy paradox. PloS one, 9(1), e84217. http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0084217 DOI:10.1371/journal.pone.0084217[1]
    • It is now acknowledged that high accuracy is not necessarily an indicator of high classifier performance and therein lies the accuracy paradox [16]–[18]. For instance, in a predictive classification setting, predictive models with a given (lower) level of accuracy may have greater predictive power than models with higher accuracy. This deleterious feature is explained in-depth in Section sec:crit-accur-using. In particular, if a single class contains most of the data, a majority classifier that assigns all input cases to this majority class (the one concentrating the probability mass of ) would produce an accurate result. Highly imbalanced or skewed training data is very commonly encountered in samples taken from natural phenomena. Moreover, the classes' distributions of the samples do not necessarily reflect the distributions in the whole population since most of the times the samples are gathered in very controlled conditions. This skewness in the data hinders the capability of statistical models to predict the behavior of the phenomena being modeled and data balancing strategies are then advisable [19].

2012

  • (Russell & Cohn, 2012) ⇒ Jesse Russell and Ronald Cohn (2012) "Accuracy Paradox", Book on Demand ISBN: 5510922974, 9785510922974 [2]
    • The accuracy paradox for predictive analytics states that predictive models with a given level of accuracy may have greater predictive power than models with higher accuracy. It may be better to avoid the accuracy metric in favor of other metrics such as precision and recall.

      Accuracy is often the starting point for analyzing the quality of a predictive model, as well as an obvious criterion for prediction. Accuracy measures the ratio of correct predictions to the total number of cases evaluated. It may seem obvious that the ratio of correct predictions to cases should be a key metric. A predictive model may have high accuracy, but be useless.

      In an example predictive model for an insurance fraud application, all cases that are predicted as high-risk by the model will be investigated. To evaluate the performance of the model, the insurance company has created a sample data set of 10,000 claims. All 10,000 cases in the validation sample have been carefully checked and it is known which cases are fraudulent. To analyze the quality of the model, the insurance uses the table of confusion. The definition of accuracy, the table of confusion for model M1Fraud, and the calculation of accuracy for model M1Fraud is shown below.