Predictive Modeling Overview

Principles

Definition

Predictive modeling is a methodological approach within statistics, machine learning, and data science focused on developing models that forecast future outcomes or estimate unknown values based on patterns identified in historical data. This field investigates the relationships and dependencies within data to construct algorithms capable of generalizing to new inputs, characterized by the application of various statistical and computational techniques for model building, validation, and evaluation. Its significance lies in enabling informed decision-making, risk analysis, optimization, and generating prospective insights across a wide range of academic disciplines and practical applications.

Ontological type

Model Evaluation

Core Methods

Feature Engineering

Key developments

Key Figures

High-Dimensional Predictive Inference era

James H. Stock [1] is associated with Harvard University [3] and Northwestern University [4] during this era. His key contributions in this high-dimensional predictive inference era come from forecasting with many predictors via dimension reduction, most notably the paper Forecasting Using Principal Components From a Large Number of Predictors [6], which shows how principal components can condense information and stabilize forecasts in large predictor sets. Mark W. Watson [2] is associated with Northwestern University [4] and Princeton University [5] during this era. Watson's contributions mirror Stock in advancing high-dimensional forecasting through the same principal-components framework, enabling probabilistic forecasts from condensed inputs and laying groundwork for later ensemble and model-uncertainty approaches as discussed in Forecasting Using Principal Components From a Large Number of Predictors [6].

Probabilistic Calibration and Validation era

Adrian E. Raftery [1], associated with the University of Washington [3] and Heidelberg University [4], emerges as a leading figure in probabilistic calibration and validation during the 2004-2017 era. His key contributions in this era include formalizing strictly proper scoring rules for prediction and estimation, as illustrated by the 2007 paper Strictly Proper Scoring Rules, Prediction, and Estimation [6], which advanced trustworthy uncertainty quantification and principled forecast evaluation. Tilmann Gneiting [2], affiliated with Stanford University [5] and the University of Washington [3], contributed to the same strand of work on strictly proper scoring rules [6], reinforcing a rigorous framework for forecast verification and calibrated distributions. Together, these contributions supported trustworthy calibration and evaluation frameworks that enabled scalable, decision-relevant probabilistic forecasting across domains.

Cross-Model Predictive Interpretability era

Marko Sarstedt [1] is associated with Ludwig-Maximilians-Universität München [3] and University of Newcastle Australia [4] during the Cross-Model Predictive Interpretability era. His key contributions in this era center on advancing predictive model assessment in PLS-SEM [7] and developing guidelines for PLSpredict [7], which helped standardize cross-model predictive interpretability across heterogeneous methods. Christian M. Ringle [2] is affiliated with James Cook University [5] and Universität Hamburg [6] during this period. His key contributions in this era focus on the same predictive model assessment and cross-model interpretability themes, notably through the 2019 Predictive model assessment in PLS-SEM: guidelines for using PLSpredict [7], which facilitated comparisons of feature effects across model classes.