Special Issue “Algorithms for Feature Selection”

1. Introduction

This Special Issue of the open access journal Algorithms is dedicated to showcasing cutting-edge research in algorithms for feature selection. My call for papers sought original articles presenting recent breakthroughs and the state of the art in various research areas related to feature selection techniques. In this context, my primary focus was on the expansive field of feature selection, where I sought exceptional papers that delved into both theoretical and practical aspects. These included evolutionary search techniques for feature selection, ensemble methods for feature selection, addressing feature selection challenges in high-dimensional and time series data, and exploring applications in textual data and deep feature selection.

The response to the call was overwhelming, with twelve research papers submitted to the Special Issue, each delving into a different facet of feature selection. Following a rigorous peer review process led by invited experts to ensure the highest standard, I am proud to present six exceptional research papers that have been accepted for publication in their revised form.

I anticipate that the contents of this issue will be of immense value not only to active researchers in the field but also to anyone with an interest in theoretical computer science. The collective expertise presented here promises to enhance our understanding of feature selection techniques, paving the way for advancements in various domains. I invite you to delve into the insightful discoveries shared within these pages and hope they inspire new avenues of exploration in the fascinating realm of algorithms for feature selection.

2. Special Issue

The first accepted paper [1] studies data mining, visualization, and predictive analysis to delve into diverse factors, including patient characteristics, disease prevalence, hygienic conditions, water quality, chemicals, therapeutic medications, and their potential side effects. The valuable findings of this study have the potential to empower government entities, healthcare practitioners, and non-governmental organizations with essential insights into waterborne diseases. Consequently, this knowledge aids the implementation of effective preventive measures, fostering enhanced public health and safety, particularly in middle- and low-income nations, and contributing positively to the global health landscape.

In the second accepted paper [2], the focus lies on addressing the challenge of mitigating the influence of irrelevant features and improving the fusion of multimodal data. The study introduces a novel approach by employing a deep autoencoder to fuse information from diverse modalities early and effectively remove noise, thereby enhancing model performance. To further enhance the fusion effect across various modalities, a multi-classifier hybrid prediction approach is used in the final phases. This paper presents a promising solution to optimize multimodal data analysis and has significant implications for advancing the field of data fusion and predictive modeling.

Paper [3] focuses on the creation of an automated method for pixel-level identification, in particular on the segmentation of deep cracks on historical surfaces. To enhance the performance of the technique, a variety of U-Net deep learning architectures are investigated, as well as the efficacy of various loss functions, such as Dice, cross-entropy (CE), and a hybrid combination. The team assembled a main dataset of crack photos from historic Cairo, Egypt, collected over a two-year period to help the research. They also used a contrast stretching technique to examine how different environmental factors affect old surface photos. To decrease the requirement for manual pixel-by-pixel annotation and increase process efficiency, a novel semi-supervised pixel-level map generating module was also developed. With greater accuracy and productivity, this thorough technique shows great potential for increasing fracture detection on ancient surfaces.

Paper [4] introduces a novel unsupervised feature selection method, comprising two distinct phases. Initially, the method constructs a dense feature subgraph by utilizing mutual information to facilitate effective feature selection. Subsequently, the dynamic clustering of features based on shared nearest neighbors is performed in the second phase. The chosen set of features is then supplemented with the resultant cluster representatives, which are mutually exclusive to the feature sub-graph. The researchers experimented on five common UCI datasets to gauge the method’s performance. They compared its effectiveness against five existing feature selection techniques, evaluating using two crucial performance parameters: accuracy (ACC) and Matthew’s correlation coefficient (MCC). This study not only contributes an innovative approach to feature selection but also provides a comprehensive evaluation of benchmark datasets, showcasing its potential for practical applications.

In paper [5], the emphasis is on finding key factors that influence the success of movie crowdfunding efforts. To find and evaluate important candidate elements, text mining and feature selection approaches such decision trees, rough set theory, and ReliefF methods were used. Support vector machines were then used to analyze these potential characteristics, which helped identify the critical elements that have a significant impact on the success of movie crowdsourcing efforts. The study’s findings provide insightful information and useful recommendations for enhancing the likelihood that the next crowdfunding campaigns in the film industry will be successful.

Paper [6] introduces innovative data preprocessing algorithms designed for feature extraction from speech messages. Two distinct approaches are proposed for this purpose. In the first method, autocorrelation functions of messages are built using the Fourier transform, but in the second method, autocorrelation portraits of voice signals are created. The given methods operate with pairs of samples from the original signal and are reasonably easy to construct; however, they need the usage of cyclic operators. The researchers used the created technology to the specific goal of deciphering phraseological radio exchange signals in the Russian language to gauge its performance. They significantly increased recognition accuracy by 1.7% by integrating the algorithm with preliminary feature extraction in comparison to earlier techniques. This result highlights the practical benefits of the proposed techniques in enhancing the analysis and recognition of speech messages. Overall, the research indicates that these preprocessing approaches show promise in improving the performance of speech analysis and recognition tasks, offering potential applications in various real-world scenarios where speech processing is essential.

The dedicated editorial staff of the journal Algorithms deserves my heartfelt gratitude as the Editor. Their outstanding commitment and cooperation throughout the publication process have been instrumental in ensuring the quality and timely dissemination of valuable research. I am also sincerely appreciative of all the referees for their meticulous and prompt evaluation of the submitted works, which has significantly contributed to maintaining the high standards of our journal. Finally, I extend my thanks to all the authors who have enriched this journal with their engaging work in various aspects of scheduling research.