ISODS - Applied Data Science & AI

Evaluation on Classifier Performance Across Varied Data Types in Public Datasets Analysis

Sang Vu, Nguyen Hoang Thong, Vo Van Tu, Nguyen Duy Tan

In the rapidly evolving field of machine learning, the performance of classifiers is often influenced by the nature of the data they are applied to. This manuscript presents a comprehensive evaluation of the performance of various machine learning classifiers—Random Forest (RF), XGBoost (XGB), k-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Artificial Neural Networks (ANN)—across three distinct data types: categorical, mixed, and numerical. Utilizing publicly available datasets and default classifier parameters, the study employs the F1-score as the key performance metric. The findings reveal that ensemble methods like RF and XGB excel in handling both categorical and mixed data types, while XGB stands out for numerical data. On the other hand, classifiers like KNN and SVM face challenges with categorical and mixed data types, respectively. Given that the study was conducted using default parameters, future work should focus on hyperparameter tuning to optimize classifier performance across different data types. This research serves as a valuable resource for machine learning practitioners, offering insights into effective model selection based on data characteristics.

Role of Advanced Data Analytics in Higher Education: Using Machine Learning Models to Predict Student Success

Harold Tobias Adu-Twum, Emmanuel Adu Sarfo, Evans Nartey

This research article explores the pressing issue of college student dropout, employing a suite of predictive modeling techniques to forecast students' likelihood of discontinuing their studies. In an era where higher education plays a crucial role in individual career prospects and societal progress, understanding and mitigating dropout rates is of paramount importance. We leverage a comprehensive dataset, incorporating demographic, socioeconomic, academic, and financial factors, to train and test predictive models: Logistic Regression, Random Forest, Decision Tree Classifier, Support Vector Machine (SVM), and Gradient Boosting. Our analysis reveals that the gradient boosting model outperforms its counterparts in predicting student dropout, achieving the highest precision, recall, and F1 score among the evaluated models. Gradient boosting model correctly identified 94.4% of the student as dropouts. This superiority underscores the gradient boosting’s robustness in handling the complex, multidimensional nature of factors influencing college retention. Moreover, it was found that circular units approved in the first semester is the most important factor in determining student success in terms of graduating or dropping out. This is then followed by tuition fees up to date. The previous qualification of students has the lowest predictive power indicating that current circumstances of students – academic courses, and finances contributes to students’ success more. The findings underscore the potential of machine learning in crafting targeted interventions to support at-risk students, thereby enhancing retention rates and contributing to the broader objectives of educational equity and success.

Vietnam University Ranking in Data Science and Artificial Intelligence through Principal Components Analysis

Linh Dinh, Nam Vu, Cuong Do, Thanh Ha, Tan Nguyen, Robert Pham

As Data Science (DS) and Artificial Intelligence (AI) disciplines are rising within the global technological sphere, students are actively seeking options that potentially enhance their knowledge of these domains. Nevertheless, the absence of a robust ranking system in Vietnam poses a considerable challenge for prospective college students aspiring to pursue careers in Data Science or Artificial Intelligence. The team is driven by the primary goal of establishing a comprehensive ranking framework to aid students in selecting universities conducive to their academic and career pursuits in DS and AI. The methodology comprises a systematic process of gathering and transforming data, conducting analytical tests such as correlation tests and comparative analysis of data points, and constructing linear regression models and principal components analysis utilizing Excel and R Studio employed for execution. Principal Components Analysis (PCA) was chosen as the main method to approach the dataset in ranking various features and concluding features that are significant in determining quality universities. The study successfully yields the first ranking system within the Vietnamese Data Science and Artificial Intelligence college sectors that captures the unique advantages and values of 36 Vietnamese universities. Notably, students, lecturers, and lawmakers who seek counsel in the dynamic field of Data Science and Artificial intelligence education can gain useful perception and development from the ranking system, exponentially making more comprehensive evaluation for future studies of Data and AI.

An Overview of Facial Emotion Recognition Using Deep Learning

Vo Nhat Anh Nguyen, Tan Duy Le, Thinh Vo Xuan, Tan Duy Le, An Mai, Kha Tu Huynh

The paper presents an overview of Facial Expression Recognition methods using deep learning models that have been reviewed from published papers in reputable journals and conferences. Based on these studies, the authors have categorized the methods into five groups, including Convolutional Neural Network, Generative Adversarial Networks, Recurrent Neural Network, Transfer Learning and Other Methods and Signal-based Methods. In each group, the authors summarize prominent methods and comparative results. From these groups of methods, the authors have suggested a general process for the Facial Expression Recognition problem consisting of six steps of Problem Statement, Image preparation, Data Pre-processing, Feature extraction, Deep learning approaches and Facial expression classification. The paper provides researchers in the field of Facial Expression Recognition or newcomers to this field with an overview as well as a basis for developing in-depth research or developing solutions and models applied to related areas.