Method for Excluding Noise Features in Applying a Learning Machine

Previous works have demonstrated that a computerized learning machine could be successfully applied to predicting the presence or absence of a certain microbial feature in a microorganism. To achieve the prediction, however, the presence or absence of many other features must be examined beforehand. This paper deals with the method for decreasing the number of characters that must be examined prior to the prediction. In the six instances studied, this number could be decreased from 93 to 11 in the most fortunate case and to 26 in the most unfortunate case, without causing deterioration of the prediction results.