Browsing by Author "Uyar,A."

Now showing 1 - 3 of 3

Citation Count: 2
Performance analysis of classification models for medical diagnostic decision support systems;
(2013) Segmen,E.; Uyar, Aslı; Bilgisayar Mühendisliği / Computer Engineering
As a part of electronic healthcare systems, medical diagnostic decision support systems have been more popular in clinical routine. It is critical to decide the best model to provide reliable machine learning based decision support in diagnostic problems. In this study, the performance of common classification algorithms have been comparatively evaluated using public medical datasets. The experimental results reveal that, although there is no single best algorithm for all datasets, MLP and Naive Bayes methods have provided relatively higher success rates. © 2013 IEEE.
Citation Count: 0
Rejection threshold optimization using 3D ROC curves: Novel findings on biomedical datasets
(Ismail Saritas, 2021) Uyar,A.; Uyar, Aslı; Bilgisayar Mühendisliği / Computer Engineering
Reject option is introduced in classification tasks to prevent potential misclassifications. Although optimization of error-reject trade-off has been widely investigated, it is shown that error rate itself is not an appropriate performance measure, when misclassification costs are unequal or class distributions are imbalanced. ROC analysis is proposed as an alternative approach to performance evaluation in terms of true positives (TP) and false positives (FP). Considering classification with reject option, we need to represent the tradeoff between TP, FP and rejection rates. In this paper, we propose 3D ROC analysis to determine the optimal rejection threshold as an analogy to decision threshold optimization in 2D ROC curves. We have demonstrated our proposed method with Naive Bayes classifier on Heart Disease dataset and validated the efficiency of the method on multiple datasets from UCI Machine Learning Repository. Our experiments reveal that classification with optimized rejection threshold significantly improves true positive rates in biomedical datasets. Furthermore, false positive rates remain the same with rejection rates below 10% on average. © 2021, Ismail Saritas. All rights reserved.
Citation Count: 0
Sampling bias in microarray data analysis: A demonstration in the field of reproductive biology
(IEEE Computer Society, 2013) Manafi,S.; Uyar, Aslı; Bener,A.; Bilgisayar Mühendisliği / Computer Engineering
The actual benefit from high-throughput microarray experiments strongly relies on elimination of all possible sources of biases during both the experimental procedure and data analysis process. Within the context of reproductive biology, microarray based transcriptomic analysis of oocyte and surrounding cumulus/granulosa cells poses significant challenges due to limited amount of samples and/or potential contaminations from adjacent cells. In this study, we investigated the effect of sampling bias on consistency of the microarray differential expression analysis in the field of reproduction. Experiments were conducted on five datasets obtained from publicly available microarray repositories. For each dataset, probe level expression values were extracted and background adjustment, inter-array quantile normalization and probe set summarization were performed according to the Robust Multi-Chip Average algorithm. Genes with a false discovery rate-corrected p value of <0.05 and |Fold Change| > 2 were considered as differentially expressed. Results demonstrate that both number of replicates and including different subsets of available samples in the analysis alter the number of differentially expressed genes. We suggest that assessment of inter-sample variance prior to differential expression analysis is an important step in microarray experiments and proper handling of that variance may require alternative normalization and/or statistical test methods. © 2013 IEEE.