Analysis of Failure Data with Missing Labels

This paper presents a new technique for the analysis of failure data when some of the labels are missing. When multiple systems are in operation, the label associated with a failure are usually given to indicate the system type or the specific system the failure belongs to. Data records in practice often suffer from missing labels. Missing labels can be partially known or completely unknown. A statistical inference procedure based on the expectation maximization algorithm is proposed to address this problem. Give the observed data, the proposed technique derives explicitly the distribution of the missing labels. The advantage of this technique is that it is a general inference procedure and is flexible to account for different parameter settings and failure rate functions. The method is applied to real case data on lift failures. It shows that the method can well handle parameter estimation in the face of missing labels.