Neural network modeling of the level of observation decision in an acute psychiatric ward

摘要: Patients in an acute psychiatric ward need to be observed with varying levels of closeness. We report a series experiments which neural networks were trained model this "level observation" decision. One hundred eighty-seven such clinical decisions used train and test the evaluated by multitrial v-fold cross-validation procedure. network modeling approach was break down decision process into four subproblems, each solved perceptron unit. This resulted hierarchical having structure that equivalent sparsely connected two-layer perceptron. Neural approaches compared nearest neighbor, linear regression, naive Bayes classifiers. The sparse most accurate shows is nonlinear, nets can more than other statistical approaches, decomposition useful methodology for design.