Predicting Nosocomial Infection by Using Data Mining Technologies - New Contributions in Information Systems and Technologies - page 182

The most used technique to evaluate the models is the Confusion Matrix, a matrix
that measures the predictions correction [11]. This matrix has four types of results:
True Positives (TP), False Positives (FP), True Negatives (TN) and False Negatives
(FN). A TP is a positive case correctly classified, a FP is a case incorrectly classified
as positive. A TN is a case correctly classified as negative and a FN result is a case
incorrectly classified as negative. With the estimation of each of these results it is
possible to apply a set of statistical measures to evaluate the quality of the results,
being the most used sensibility, specificity and acuity [12].
Sensitivity is the capacity of the model detecting the occurrence of an event when
present and it measures the proportion between the number of TP results and all the
positive results, i.e., ்௉

்௉ାிே .
Specificity is the capacity of the model classify correctly the non occurrence of an
event and it is the ratio between the number of TN results and all the negative values,
i.e., ்ே

்ேାி௉ .
Acuity is the agreement between the correctly detected values and the real values
and it is the proportion between all the true results and all the cases, i.e.,
்௉ା்ே
்௉ାி௉ା்ேାிே .
The quality of the models obtained with this work was evaluated with these
statistical measures. In the case of this work, the sensitivity was considered by
predicting the non-occurrence of an infection and the specificity was considered the
capacity of predicting the occurrence of an infection. Some of the models allowed to
achieve the best overall four results for each one of the DM techniques used
(Table 2). The best models were selected considering the values of sensitivity because
it is important to identify all the non-occurrences of infection. Knowing which are the
group of non-occurrence of infection, it is possible consider all the other predictions
as risk groups capable of acquiring a nosocomial infection.

Table 2. Top 4 models for each DM technique

Support Vector Machine Naïve Bayes

Specificity Sensitivity Acuity Specificity Sensitivity Acuity

Scenario 1 and
Approach B

0.763 0.919 0.838 0.733 0.941 0.825

Scenario 2 and
Approach B

0.741 0.942 0.831 0.733 0.941 0.825

Scenario 1 and
Approach C

0.731 0.793 0.766 0.733 0.941 0.825

Scenario 3 and
Approach C

0.675 0.845 0.754 0.733 0.941 0.825

3.7 Deployment

After the evaluation of the models, the knowledge obtained can be used by healthcare
professionals to predict the occurrence of an infection in the presence of the studied
risk factors. The combination of variables that have a higher probability to result in