Clean and safe water is essential for human health and recognized as a fundamental right. However, the degradation of water quality worldwide, due to pollution and inadequate water management, poses serious risks to public health, making it crucial to evaluate water potability effectively. This study aims to classify water potability more accurately by leveraging machine learning models. The research utilized the Water Potability Dataset, which contains 3,276 records and 10 parameters, such as pH, hardness, and solids, all of which were analyzed to ensure a comprehensive assessment of water quality. Missing values in the dataset were addressed through imputation techniques, while class imbalance was resolved with data balancing methods. The dataset was divided into 80% for training and 20% for testing, ensuring reliable model validation. Decision Tree and Random Forest models were applied due to their efficiency in managing complex datasets and providing interpretable outputs. The Random Forest model outperformed the Decision Tree, achieving higher accuracy and demonstrating superior reliability in classifying water as potable or non-potable. Evaluation results indicate that these models effectively classify water as potable or non-potable, showcasing the potential of machine learning to improve water quality analysis. The findings of this study provide a reliable, data-driven framework that can support policymakers, water management authorities, and public health organizations in addressing water quality challenges. By facilitating timely and accurate assessment, this research contributes to safeguarding public health and promoting sustainable water management practices.
This study involved designing and implementing a compact educational blockchain system that facilitates understanding by eliminating much of the complexity of blockchain technology. The system, implemented in Python, ensures that the concept of blockchain integrity is easy to comprehend. Essential features include operation of consensus algorithms and use of block generation/addition to facilitate learning. The effectiveness of the system was evaluated via a survey of university students. The results revealed significant improvements in blockchain comprehension. These findings contribute to the development of foundational systems for blockchain education in academic settings.
Lung cancer remains to be a major cause of cancer-related mortality worldwide, that underscores the necessity of an early and precise diagnosis. This study investigates how feature selection can improve machine learning models for the diagnosis of lung cancer. Utilizing the Lung Cancer Dataset, we employed ranking-based feature selection methods, specifically Information Gain and Gini Decrease, for determining the most important features for model training. Three classifiers—Random Forest, Decision Tree, and Support Vector Machine (SVM)—were evaluated using measures including AUC, Classification Accuracy, F1 Score, Precision, Recall, and MCC. The results demonstrated notable improvements in model performance with feature selection, highlighting its importance in reducing computational complexity and improving diagnostic accuracy. The Random Forest model demonstrated the most favorable AUC performance of 0.980 and an accuracy of 94.4%, underscoring its robustness in predicting lung cancer outcomes. This study emphasizes the value of integrating feature selection into machine learning workflows to build efficient, interpretable, and clinically applicable models for lung cancer diagnosis.
Cricket is one of the most famous sports in the world that involves no competition alone, but a lively community of players, supporters, coaches, and technology development. There is a lack of automated systems offering useful, personalized feedback to enable the players to improve their techniques in trying to imitate their idols. This paper presents a data-agnostic framework that integrates data visualization, machine learning, and computer vision to allow users to analyze and emulate various stances of professional cricket players. Results obtained using a carefully curated dataset of classic cricket shots, including famous pull shots and cover drives, demonstrate how well the presented framework analyzes and contrasts batting strategies. It generates detailed results comprising a similarity comparison score and a two-dimensional visualization of posture differences, thus enabling users to further refine their techniques. Designed initially for cricket, the framework’s flexibility bridges the gap in emulating idols and allows for adaptation to other sports, demonstrating its broad applicability and innovative approach.
This paper presents a novel forest fire inventory dataset for Peninsular Malaysia, created using public remote sensing data spanning from 2001 to 2023. The dataset leverages the Google Earth Engine (GEE) framework established in our previous study, designed to address the lack of comprehensive, publicly accessible fire dataset for this region. It includes a total of 7349 columns (attributes) and 11,279 rows, consisting of 5650 fire points and 5629 non-fire points. The attributes encompass annual, seasonal, and monthly dynamic variables, as well as static, administrative, temporal variables, and class labels. As a result of this research, we have published and made the dataset publicly accessible at https://doi.org/10.5281/zenodo.11542164. This dataset is aspired to serve as a valuable resource for researchers and domain experts, facilitating further analysis and deeper understanding into forest fire behaviour in Peninsular Malaysia.
