Supervised Classification, Machine Learning, and AI
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
The generic form of supervised classification problems is described, and attention is drawn to the importance of independence of the training and test sets in performance evaluation. Some ways in which problems can arise because of differences between the training and test set distributions are described, including non-stationarity (or population drift), clinical trial inclusion criteria, and general challenges of selection bias and reject inference. The need for careful consideration of the objective is illustrated by contrasting diagnosis and screening. The challenge of measuring classifier performance, and the need for careful thought about the aims when trying to answer the question “which classifier is best?” is discussed. A distinction is drawn between non-accuracy aspects of performance (such as speed of classification, handling missing data, the ability to explain decisions, and ease of use) and accuracy-based measures. The latter are explored in detail, covering the basic confusion matrix, error rate, sensitivity (recall), specificity, precision (positive predictive value) and negative predictive value, the Area and Partial Area Under the Curve, the H-measure, false discovery rate, the F-measure, and others.
Fields of Study & Subjects
Imbalanced Data Classification Techniques
Machine Learning and Data Classification
Artificial Intelligence in Healthcare
Publication details
- Journal / source
- Chapman & Hall/CRC eBooks
- Pages
- 171–185
- Publication date
- 27 March 2026
- Language
- EN
- Document type
- Book chapter