Modelling of the Feed Forward Neural Network with its Application in Medical Diagnosis

This study explores a data mining technique to solve the problem associated with the medical diagnosis of acute
inflammations and acute nephritises of the urinary system. Medical diagnosis is a complex classification
problem that lacks an analytic or algorithmic solution. The diagnosis problem considered is a classification
problem with two types of decision patterns. First, the pattern of the problem is irregular and therefore hard to
explicitly derive an analytical or algorithmic solution. Second, the pattern is to determine the classification
accuracy reached. Modelling of the data was done using Back Propagation. A feed forward neural network
model using six neuron input and varied numbers of hidden neuron was used. The model is trained and tested by
partitioning the data into a ratio of four to one (4:1). The four-fifth, which is eighty percent (80%) of the data, is
then used for training while the remaining one-fifth, twenty percent (20%), is used for testing the trained neural
network model. The output is compared to a known result until the network output error is significantly reduced.
The output shows the classification accuracy of the model to be approximately ninety percent (90%), which
implies that only one out of ten classifications is incorrect.