Agbator, L. O

Agbi, A

Abstract

Feature selection algorithms lie at the heart of machine learning, playing a crucial role in identifying and prioritizing essential attributes within datasets. By refining sample attributes, these algorithms aim to elevate classification performance and pinpoint the most relevant features associated with distinct data classes. The primary objective is to optimize classification and prediction by selecting features based on their effectiveness. This study introduces a novel feature selection algorithm, Nonnegative Discriminative Feature Selection (NDFS), incorporating spectral analysis and multi-variate regression into stock price trend prediction. The chosen characteristics using NDFS are evaluated for classification prowess with three different classifiers. Furthermore, the performance of NDFS is benchmarked against three feature selection algorithms from existing literature. The findings indicate that NDFS emerges as a competitive technique capable of improving the efficiency of machine learning algorithms, particularly in stock price trend prediction. The adopted feature selection method performed well, achieving an accuracy of 85% in the Naïve Bayes model and an F1-score of 83%, indicating a balanced measure of precision and recall. The receiver operating characteristics curve reveals an optimal model with performance surpassing 80% when utilizing features selected by Nonnegative Discriminative Feature Selection.

   

Full-Text:

---

   

Published: 2025

---

   

Issue:

Articles

---