Fig. 9

ROC curve of different classifiers for the (a) breast and (b) lung cancer datasets: (a) In the breast cancer dataset LABS-HDMR uses RQDA to infer log-likelihood ratio of partial observations, greatly improves RQDA’s accuracy, and is followed by GLMs. LAS-HDMR closely follows GLM and RLDA, uses RQDA in its design, improves RQDA’s accuracy for small false positives, and performs similar to RQDA for large false alarms. (b) In the lung cancer dataset GLM performs best, but is closely followed by LAS-HDMR and LABS-HDMR, which both improve on the AUC of the intrinsic machinery they use for computing log-likelihood ratio of partial observations