Feature selection using metaheuristics made easy: Open source MAFESE library in Python

Artificial intelligence (AI) often relies on feature selection (FS) to recognize and highlight the most relevant and major features in a dataset. The procedure of training and optimizing AI systems with key data points is decisive for its development and efficacy. To address this challenge, the present study introduces MAFESE, an open-source Python library that employs metaheuristic algorithms for selecting the optimal set of attributes, particularly when dealing with complex and high-dimensional data. MAFESE encompasses a wide range of feature selection techniques, including unsupervised-based, filter-based, embedded-based, and wrapper-based methods. Notably, within the wrapper-based category, MAFESE offers users access to over 200 metaheuristic algorithms, empowering them to choose the most suitable algorithm based on their specific datasets and requirements. Additionally, MAFESE incorporates built-in evaluation metrics that enable efficient comparison among various algorithms. Open-source design of MAFESE encourages cooperation within the data science community, allowing for continuous upgrades. This collaborative environment promotes the sharing of ideas, proposals, and changes, resulting in a stronger and more adaptive feature selection framework. MAFESE distinguishes itself with an easy-to-use Python interface that follows object-oriented programming concepts. It supports both experienced researchers and practitioners. MAFESE offers many resources, including documentation, examples, and test cases, for a smooth user onboarding experience. The modular architecture enables users to enhance features and interface with other tools, such as scikit-learn. MAFESE can be a valuable tool for identifying meaningful features in complicated, high-dimensional datasets, making it a significant addition to feature selection. MAFESE utilizes metaheuristic algorithms to help users tackle complex feature selection problems more efficiently and accurately. Through openness and teamwork, MAFESE may become a comprehensive resource that responds to the evolving demands of the data science community. The source code and materials of MAFESE are publicly available in the official GitHub repository at https://github.com/thieu1995/mafese.

If you would like to get a version of the paper, let send me an email.

Recommended citation:
Download Paper | Download Slides