Supervised Machine Learning: Optimization Framework and Applications with SAS and R

AI framework addresses bias-variance tradeoff for supervised learning methods in real-life applications. It uses bootstrapping, statistical experiments, and data contamination to build robust classifiers. The framework is available on GitHub.

Format: Paperback / softback
Length: 182 pages
Publication date: 29 April 2022
Publisher: Taylor & Francis Ltd

Supervised learning methods in real-life applications often face a tradeoff between bias and variance. To address this challenge, a comprehensive AI framework has been developed. This framework encompasses several key features, including:

Bootstrapping: The process of creating multiple training and testing data sets with varying characteristics is known as bootstrapping. This helps in dealing with bias by ensuring that the classifiers are not overfitting the data.

Design and Analysis of Statistical Experiments: Statistical experiments are used to identify optimal feature subsets and hyper-parameters for ML methods. By analyzing the data, researchers can determine the features that contribute most to the classification task and optimize the hyper-parameters to improve the performance of the classifiers.

Data Contamination: Data contamination is a technique used to test the robustness of the classifiers. It involves introducing noise or perturbations to the training data to see how the classifiers perform under different conditions. This helps in identifying any biases or limitations of the classifiers.

The AI framework is designed to be a table-driven environment, allowing for easy management of all meta-data related to the proposed framework. It also provides interoperability with R libraries, enabling a wide range of statistical and machine-learning tasks to be accomplished.

To facilitate the development and usage of the AI framework, computer programs in R and SAS have been made available on GitHub. These programs enable researchers to create and customize the framework to suit their specific needs and integrate it into their research projects.

In conclusion, the AI framework developed in this study offers a comprehensive approach to addressing the bias-variance tradeoff in supervised learning methods. By leveraging bootstrapping, statistical experiments, and data contamination techniques, researchers can build more robust and generalizable classifiers that can be applied to real-life applications. The availability of the framework on GitHub makes it accessible to a wide range of researchers and practitioners, fostering further advancements in the field of AI.

Dimension: 234 x 156 (mm)
ISBN-13: 9780367538828