Classification and Clustering for predicting breathalyzer failures.

This paper reports the use of Machine Learning, classification and clustering to predict failures of breathalyzers used in traffic enforcement. Classification is based on training with data from periodic verifications, where errors and measurement deviations are analyzed against a certified reference material. Clustering finds groups based on similarities, according to the responses of electrochemical cells built into the instruments. Both approaches are valuable tools for monitoring breathalyzer wear, and their applications reduce the occurrence of false-positives and false-negatives in field sobriety checkpoints. The choice of one of the two or a combination of both can be applied by regulatory bodies or companies that carry out repairs, in order to provide greater assurance of quality in the use of the breathalyzers.