23.3 Efficient PFA usage For the best use of Predictive Failure Analysis (PFA), perform these actions: Reduce the number of false positives. To avoid false positives, PFA can perform "supervised" learning, which excludes certain data that PFA uses when making predictions of future behavior. For example, when you suspect certain jobs or address spaces are inconsistent and have the potential of being restarted often, you can increase the accuracy of PFA by excluding the job or address spaces from analysis. Supervised learning applies to the following checks: ­ ­ ­ ­ "LOGREC arrival rate check" on page 501 "Frames and slots usage check" on page 501 "Message arrival rate check" on page 501 "SMF arrival rate check" on page 502 Eliminate the jobs causing false positives. The supervised learning service can help you avoid false positives by excluding certain data that PFA uses when making predictions of future behavior. To minimize the impact to check performance, only use EXCLUDED_JOBS for the conditions that cause you the most inconvenience. Instead, use other tuning parameters for the check such as STDDEV. A sample EXCLUDED_JOBS file ships in the /usr/lpp/bcp/samples/PFA directory. It is named EXCLUDED_JOBS and includes an example comment line. You can modify the file using the OEDIT command, and then use the F PFA,UPDATE command to have PFA read in the contents of the file and start to use it in during processing. See 23.3.2, "Eliminate jobs causing false positives" on page 493, for more information