ZTZerin Akter Tanni

Zerin Akter Tanni

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Zerin Akter Tanni has a total of 10 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Probabilistic Machine Learning Models for Uncertainty-Aware Predictions

    by Harleen Kaur, Janardhana Naidu Kola, Prathamesh Sawarkar, Shubham Malhotra, Ponnarasan Krishnan, Zerin Akter Tanni

  • Open Access

    Diagnostic Accuracy and Probability Reliability of Data Science Models for Breast Cancer Cytology

    by Mohammed Majid Bakhsh, Md Yousuf Biplob, Tushar Roy, Zerin Akter Tanni, Md Nazmussakib, Mahmudul Haque Rijvi

    Data science can support cancer diagnosis, but strong classification results do not establish that a model provides reliable probabilities or can improve patient care. Breast cancer cytology offers a useful setting for examining these distinctions because public nuclear measurements allow transparent computational experiments. Comparing diagnostic discrimination, probability accuracy, and classification errors among three established models using a common validation design. An original secondary analysis was conducted on the Wisconsin Diagnostic Breast Cancer dataset, comprising 569 observations, including 212 malignant and 357 benign diagnoses, with 30 nuclear measurements. Penalized logistic regression, a radial basis function support vector machine, and a random forest were evaluated through fivefold outer cross-validation with threefold inner hyperparameter selection. Scaling was fitted within training pipelines. The primary partition used seed 42; two additional partitions examined split sensitivity. Outcomes included area under the receiver operating characteristic curve (AUROC), average precision, Brier score, log loss, and sensitivity and specificity at a fixed probability threshold of 0.50. Stratified bootstrap intervals were calculated from the primary out-of-fold predictions. The support vector machine achieved an AUROC of 0.9949, compared with 0.9934 for logistic regression and 0.9899 for the random forest. Its Brier score was 0.0189, compared with 0.0228 and 0.0335, respectively. At the 0.50 threshold, the models missed 10, 12, and 15 malignant observations. The paired AUROC difference between the support vector machine and logistic regression was 0.0015, with a conditional 95% bootstrap interval from -0.0005 to 0.0041. The support vector machine retained the lowest Brier score in both additional partitions. Established models separated benign and malignant cytology observations well in this benchmark. The support vector machine had the most favorable probability accuracy, while logistic regression remained competitive. Small differences in discrimination, persistent false negatives, and limited population information prevent conclusions about clinical superiority. Independent contemporary validation is required before considering patient-facing use.