FEATURE ENGINEERING IN MACHINE LEARNING : SELECTION, EXTRACTION, AND THEIR IMPACT ON MODEL PERFORMANCE

Abstract

In the data pre-processing stage of machine learning (ML), feature selection and extraction are  essential procedures. The models' computational efficiency, interpretability, and performance are all  greatly impacted by these stages. Finding the most relevant characteristics in a dataset while  maintaining important information is known as feature selection. However, in order to facilitate more  efficient learning, feature extraction converts unstructured data into meaningful representations. The  present work offers a thorough review of feature selection and extraction, emphasizing its  significance, different approaches, and applications in a range of fields. The research also highlights  issues like the Curse of Dimensionality, the existence of distracting or noisy characteristics, and the  efficacy being limited by the trade-off between interpretability and performance. In order to improve  scalability and cross-domain adaptation, the research finishes with insights into new trends and  potential paths, such as hybrid techniques and the combination of feature selection with deep  learning.