Скачать или смотреть Segmented Regression (Alternate algorithm)

This video presents the development of an alternate algorithm designed to identify optimal segmented regression models, focusing on the associated probabilities. Segmented regression, essential in fields like economics and biology, models datasets with distinct linear segments. The algorithm systematically excludes subsets of data points, generating and evaluating numerous potential models to find the one that minimises prediction error for the excluded data.
To address the complexity of generating many models, the video introduces a probabilistic framework that compares models based on their predictive performance. Several probabilistic metrics are proposed to assess the likelihood of one model being superior to another. This approach is especially valuable in cases with clear data breakpoints, allowing for more efficient model selection by concentrating on significant segments.
It concludes by discussing how combining these probabilistic conditions can improve the accuracy and reliability of segmented regression models, offering an alternate approach to handle datasets.