This document serves as an expanding comparative index mapping alternative regression families used across machine learning. It spans classical straight-line estimators and non-linear, non-parametric tree partitioning architectures.
Parametric modeling baselines utilizing straight ordinary least squares estimation lines to project a continuous scalar target variable.
A fundamental type of linear regression mapping exactly one explanatory or independent variable to predict a single continuous scalar response variable.
An expanded type of linear regression leveraging two or more independent input features simultaneously to approximate the continuous scale variance of a single scalar target variable.
Advanced algorithmic variants optimized to capture non-linear structures, conditional feature partitions, and multi-variable interaction boundaries.
A type of regression modeling curvilinear relationships by raising independent input parameters to distinct mathematical powers. It is widely used to capture non-linear tracking dynamics, such as modeling how diseases spread, tracking pandemics, or forecasting epidemic velocity over time.
An advanced algorithmic framework that utilizes a strict margin of tolerance (the epsilon-tube) wrapped around a centerline. The optimization engine focuses on capturing as many empirical data coordinates as possible inside this margin width while ignoring smaller residual variations.
A non-parametric machine learning algorithm that recursively implements explicit rule-based splits to partition multi-dimensional data plots into uniform, localized neighborhood subsets based on variance minimization benchmarks.
A robust version of ensemble machine learning utilizing bagging mechanics. It generates extensive collections of deep, un-pruned decision trees, building individual models by picking random data rows and random feature subsets from the primary training block, before aggregating their predictions into a unified final output.