20+ ML algorithms listed

Machine learning is software that learns patterns from examples, and the most useful first split is simple: some algorithms predict a labeled answer, while others group unlabeled data by similarity. (scikit-learn.org) Scikit-learn’s user guide groups common methods into supervised learning, including linear models, support vector machines, nearest neighbors, decision trees and neural networks, and unsupervised learning, including clustering and dimensionality reduction. Google’s Machine Learning Crash Course teaches the same workflow through classification, regression and clustering lessons. (scikit-learn.org) (developers.google.com) A list with 20-plus algorithms is really a map of tradeoffs. Linear and logistic models are fast baselines for prices, churn and spam; k-nearest neighbors compares a new case with similar past cases; support vector machines draw a boundary between classes; naive Bayes works well when simple probability rules are enough. (scikit-learn.org) Tree methods answer questions by splitting data into branches, like a flowchart for “yes” or “no.” A single decision tree is easy to inspect, random forests average many trees trained on resampled data, and gradient boosting adds trees one after another to fix earlier mistakes. (scikit-learn.org) (xgboost.readthedocs.io) That last family matters because boosted trees still dominate many structured-data jobs such as credit risk, fraud scoring and demand forecasting. XGBoost describes itself as an optimized gradient boosting library, and recent tabular-data benchmarking papers still find tree-based models ahead of neural networks on many medium-size datasets. (xgboost.readthedocs.io) (arxiv.org) Neural networks work differently: they stack layers of weighted connections and adjust those weights during training. They are the default choice for images, audio and language, but OpenML’s benchmark documentation still treats them as one family among several standard machine-learning options rather than a universal winner. (openml.org) Other algorithms solve narrower problems. K-means clustering groups similar points without labels, principal component analysis compresses many columns into a few summary directions, and anomaly-detection methods flag rare cases that do not look like the rest of the data. (scikit-learn.org) The practical lesson in any long algorithm list is not to memorize 20 names. It is to match the tool to the data, start with a baseline you can explain, and move to heavier models only when the error, speed or scale justifies it. (developers.google.com)