Predicting Nosocomial Infection by Using Data Mining Technologies - New Contributions in Information Systems and Technologies - page 181

• Absence of Risk Factors (Scenario 1): all of the variables were used in the model

except the variable Risk Factors, i.e., the variables Age or Age Class, Sex,
Hospitalization Days, Clinical Specialty, Intubation, Catheterization and the
target variable Nosocomial Infection were used;

• Absence of Intubation (Scenario 2): all of the variables were used in the model

except the variable Intubation, i.e., the variables Age or Age Class, Sex,
Hospitalization Days, Clinical Specialty, Risk Factors, Catheterization and the
target variable Nosocomial Infection were used;

• Absence of Catheterization (Scenario 3): all of the variables were used in the

model except the variable Catheterization, i.e., the variables Age or Age Class,
Sex, Hospitalization Days, Clinical Specialty, Risk Factors, Intubation and the
target variable Nosocomial Infection were used;

All the variables (Scenario 4): all of the variables were used in the model,
i.e., the variables Age or Age Class, Sex, Hospitalization Days, Clinical
Specialty, Risk Factors, Intubation, Catheterization and the target variable
Nosocomial Infection were used.
These four scenarios were induced according to the three different datasets
previously created (Approach A, Approach B and Approach C). The DM techniques
were then applied to all combinations of scenarios and approaches (situations), in
order to create new knowledge and obtaining the best model for solving the problem.
So, the four scenarios were induced by the two DM classification techniques, the
three approaches and a target variable, which means that 24 DM models were
induced. The models generated analyze the relationship between the variables
considered and their impact in the occurrence of a nosocomial infection. These
models can be represented by the following expression:

•

M n ≡ <A f , S i , TDM y >
According to this expression, the Predictive Model n ( M n ) belongs to the DM
Approach f ( A f ) and it is composed by the Situation i ( S i ) and a DM Technique y
( TDM y ). The models are presented in Table 1. At this stage the dataset were split in
test (30%) and trainning data (the remaining 70%).

Table 1. DM models induced for all the Approaches ( A f ) and Situations ( S i )

A f א {Classification}
TDM y א {SVM, NB}

A f א {Classification}
TDM y א {SVM, NB}

A f א {Classification}
TDM y א {SVM, NB}

S i א {Scenario 1 and Approach
A, …, Scenario 4 and
Approach A}

S i א {Scenario 1 and
Approach B, …, Scenario 4
and Approach B}

S i א {Scenario 1 and Approach
C, …, Scenario 4 and Approach
C}

3.6 Evaluation

As a result of the DM process several models were obtained. These models must be
evaluated in order to check their quality and, therefore, chose the one that allows to
obtain the best results, according to the goals of the study.