An empirical analysis of Brazilian courts law documents using learning techniques

This paper describes a survey on investigating judicial data to find patterns and relations between crime attributes and corresponding decisions made by courts, aiming to find import directions that interpretation of the law might be taking. We have developed an initial methodology and experimentation to look for behaviour patterns to build judicial sentences in the scope of Brazilian criminal courts and achieved results related to important trends in decision making. Neural networks-based techniques were applied for classification and pattern recognition, based on Multi-Layer Perceptron and Radial-basis Functions, associated with data organisation techniques and behavioral modalities extraction.