Computational Methods of Feature Selection (Chapman & Hall/Crc Data Mining and Knowledge Discovery Series)
A study of cross-validation and bootstrap for accuracy estimation and model selection
IJCAI'95 Proceedings of the 14th international joint conference on Artificial intelligence - Volume 2
| Hi-index | 0.00 |
In hospital internment context, patients suspect to have pulmonary tuberculosis, especially those which have a higher transmission risk, should be allocated in isolation respiratory rooms in order to reduce infection risk. Due to high implementation costs, these rooms are only available in a restricted quantity and have to be shared with patients having other infectious diseases. Currently applied criteria have been resulting in a large number of unnecessary patient isolations. This work proposes a decision support system based on neural networks to guide patient respiratory isolation. The system identifies the probability of a patient to have pulmonary tuberculosis and was developed using medical records data from 290 Pulmonary TB suspect patients who were admitted to isolation rooms in an AIDS/TB reference hospital (IDT-HUCFF-UFRJ). Based on 26 clinical-radiological variables, the system achieved a sensitivity of 94% and specificity of 89%. This system should be validated in other settings in order to confirm this high performance and its usefulness by avoiding unnecessary patient isolation as providing an earlier diagnosis, which may reduce the contamination risk.