Pattern recognition of occupational cancer using neural networks

An application of multilayered neural networks to occupational epidemiology for the pulp and paper industry is presented. Various architectures of feedforward networks with and without hidden layers have been tested to examine the relationships between occupational exposure and cancer. The inputs to the networks consist of chemical exposures derived from epidemiological studies. The outputs are the cancer types of the patients. The results of the classification performances demonstrate that an appropriate network architecture with some preprocessing of the exposures might lead to more efficient results.<>