Had-oops!
This page is supplementary to the paper entitled "Less is more: Temporal fault predictive performance over multiple Hadoop releases" ([PDF] [BIB]), which is currently to appear in the proceedings of the 5th International Symposium on Search-Based Software Engineering (SSBSE2014).
In the paper, we investigate search based fault prediction over time based on 8 consecutive Hadoop versions, aiming to analyse the impact of chronology on fault prediction performance. Our results confound the assumption, implicit in previous work, that additional information from historical versions improves prediction; though G-mean tends to improve, Recall can be reduced.