Generally, hospitals and health centers have a specialized area called Intensive Care Unit (ICU) aimed to provide treatment for people who are extremely ill. An ICU is staffed with highly trained professionals who operate advance equipment. Patients are referred to ICUs when they have a condition that cannot be treated in a regular hospital room. For example: after being in a serious car accident, after having a heart attack or stroke, when suffering a serious infection such as pneumonia, or after having major surgery. The present research focusses in the development of prediction models based on machine learning algorithms to help planning and allocating medical staff at an ICU.

The research was carried out following a classic 4-phase methodology: analysis, design, development, and validation. In the analysis, independent variables such as patient’s age, gender, reason for visit, and arrival mode are identified. Data are then pre-processed and prepared for the following phases. Specialize software packages for data mining and machine learning were selected for the investigation. Most of the investigation’s activities were planned at this point. During the design, different subsets of the independent variables were selected to be build prediction models. The size of the training and test set as well as the size of the validation set were also defined here. The phases of development and validation were carried out entirely using the software WEKA 3.9.6. By means of executing a battery of experiment and “trial and error” adjustments the prediction models were completed.

To accomplish the goals of this research, thousands of records collected from a public hospital were considered. The dataset was divided in a part for training and test (80%) and another part for validation (20%). The approach to predict the medical specialty require for the arriving patients considered the development of models based on artificial neural networks and decision trees. In all cases, instead of a simple hold-out, a 10-fold cross-validation scheme was applied.

Results showed that, in general, prediction models with fewer classes generate higher rates of correct predictions and, therefore, they can be more useful when planning and allocating medical staff at an ICU. Even though correct prediction rates fluctuate significantly from one model to another, between 80% and 83%, some of the proposed models are reliable.

In conclusion, results showed that prediction models based on machine learning algorithms can help planning and managing the need for professionals and support staff at ICU. The research might help understand the benefits of using machine learning in hospitals and health centers.