Seeking signals in data bases using neural networks

Abstract

This paper presents a strategy to allow searching a database using information that is not originally in the database. A backpropagation neural network is proposed that learns correlations between data items in the database and generates predictions. This network technique is applied to the problem of predicting housing code violations, given census data, and other problem property data. A problem is discovered in determining the categorization of the data, and a possible solution is proposed.