Machine learning learns patterns or predictive relationships from data instead of requiring people to specify every rule. In supervised learning, labeled examples train classifiers for categories or regressors for numeric values; in unsupervised learning, data has no labels. section 1.2
Machine learning supports several kinds of analysis:
- Classification predicts class labels, such as cat, dog, or hamster, from labeled images. section 1.2
- Regression predicts numeric values from labeled records or other data. section 1.2
- Unsupervised methods analyze unlabeled data; in outlier detection, they can identify unusual records across multiple features. section 1.2 section 3.0
Traditional machine learning extends analysis beyond single-value statistical summaries to tabular and time-series data. Multivariate methods can expose subtle cases such as fraud, machine failure, network threats, or unusual astronomical observations. The broader ecosystem also includes deep neural-network methods for vision, text, and audio. section 1.5 section 3.0
Practical workflow
A Python workflow can use NumPy and pandas for data preparation, then apply established models from scikit-learn or PyOD. Effective practice compares approaches, evaluates detections, examines explanations, and treats flagged records as candidates for investigation rather than automatically meaningful findings. section 1.6
