2.5. Performance measures

A confusion matrix that contains the actual outcome and predicted outcome is used to evaluate the performance of AD classification. Table 3 presents the confusion matrix for AD classification with two outcomes. TP is the number of AD patients that are correctly classified as AD. FP is the number of NDC participants that are diagnosed as AD. FN is the number of AD patients that are incorrectly classified as NDC. TN is the number of NDC participants that are classified correctly. We use the following four measures to evaluate DELearning as well as all the base classifiers. The statistically significant comparison of performances of DELearning and benchmarks are performed by using McNemar’s test [46]. The adjustment for multiple comparisons is performed using Bonferroni correction [47].

Confusion matrix for AD prediction.

Accuracy is the proportion of all participants that are correctly classified as either AD or NDC. It is formulated as the following:

Precision denotes the proportion of predicted AD cases that are real AD patients.

The recall is the proportion of AD patients that are correctly classified. It reflects the ability of a classifier to recognize positive examples. In a medical context, recall is regarded as a more primary measure than precision [48], as the aim is to identify all real positive cases.

F1-measure provides a way to combine precision and recall into a single measure with no imbalanced manner, which can be formulated as follows: