Feature Selection Techniques
Natasa Kleanthous, Abir Jaafar Hussain · 2024
Following the exploration of data preprocessing and feature extraction in Chapter 5 , Chapter 6 progresses into feature selection techniques. This chapter stands at the crucial intersection where we refine the raw power of extensive datasets into meaningful insights, focusing on accelerometer data derived from sheep. Feature selection forms the backbone of effective activity recognition. The vast array of features extracted from datasets often encompasses redundant, irrelevant, or potentially misleading information. Such inappropriate data can negatively impact the efficiency and accuracy of predictive or classification models. While exhaustive search algorithms have their advantage in locating distinct features, their practical use faces significant challenges, especially within large, highdimensional datasets. To avoid these challenges, an array of feature selection techniques has been employed across various fields to achieve an optimal set of features. These refined features are then utilized in classification or predictive models, improving their performance and interpretability. In this chapter, we will investigate the most commonly used algorithms for feature selection, categorized into filter, wrapper, and hybrid methods. Each category offers unique approaches and advantages in feature selection, and their detailed exploration will provide a comprehensive understanding of their applications and effectiveness. We will also present a hands-on Python example using a preprocessed dataset data collected from farm animals. This example will illustrate the practical application of these feature selection techniques. Through this example, readers will gain practical experience and insights into applying these techniques to similar datasets in their respective fields.