How can we efficiently extract features for data classification? This article presents a new supervised linear feature extraction algorithm based on multivariate decision trees, aiming to reduce the computational cost associated with exhaustive searches for optimal feature subsets. Unlike unsupervised methods, supervised feature extraction leverages class labels to evaluate the quality of extracted features, improving efficiency of classifiers. The proposed algorithm adopts a wrapper model method, inducing new classifiers to evaluate each new subset of features. The goal is to decrease computation time required to induce new classifiers required to evaluate every new subset of features. The algorithm’s performance is assessed through tests with real-world data. These findings show that this new approach's fundamental value lies in its ability to significantly reduce computational time required to extract features from large databases. This has substantial implications for managing and analyzing large datasets in machine learning and data mining.
This paper, published in the International Journal on Artificial Intelligence Tools, directly aligns with the journal's focus on innovative AI techniques and tools. The proposed algorithm for supervised linear feature extraction contributes to the journal's exploration of efficient methods for data classification and machine learning.