International Journal for Applied Information Management

Enhancing Minority Class Prediction in Wearable Sensor-Based Activity Recognition Using SMOTE Oversampling

Sarmini, Chyntia Raras Ajeng Widiawati, Ika Romadoni Yunita — Tue, 01 Apr 2025 00:00:00 +0000

Wearable sensor-based activity recognition has become increasingly important in various domains, particularly healthcare and sports. However, a significant challenge in this field is the issue of class imbalance, where minority activity classes are underrepresented compared to majority classes in datasets. This imbalance leads to biased classifiers that struggle to accurately identify rare but critical activities, which is especially problematic in health monitoring scenarios. This study evaluates the effectiveness of the Synthetic Minority Oversampling Technique (SMOTE) to address class imbalance in the mHealth dataset, which contains multi-sensor data from wearable devices placed on the chest, left ankle, and right lower arm. We employ the XGBoost classifier combined with SMOTE oversampling to improve recognition performance for minority classes. Model evaluation is conducted using precision, recall, F1-score, Area Under the Precision-Recall Curve (AUC-PR), ROC curve, and calibration analysis. The results demonstrate that applying SMOTE improves minority class recall from 0.75 to 0.85 and F1-score from 0.796 to 0.865, despite a slight decrease in overall accuracy from 97% to 96.5%. The AUC-PR also increases from 0.81 to 0.88, indicating a better balance in detecting minority and majority classes. Calibration curves reveal that probability estimates still require refinement to be more reliable for decision-making. This study confirms the efficacy of SMOTE in mitigating class imbalance in wearable sensor-based activity recognition and provides valuable insights for developing more accurate and fair health monitoring systems.

Predicting IMDb Ratings of One Piece Episodes Using Regression Models Based on Narrative and Popularity Features

Hery, Calandra Haryani — Tue, 01 Apr 2025 00:00:00 +0000

This study explores the predictive modeling of IMDb ratings for episodes of the anime One Piece using a linear regression approach grounded in narrative and popularity-based features. The dataset comprises 1,122 episodes, with features including story arcs, episode types, and the number of viewer votes. After one-hot encoding categorical variables and training the model using Ordinary Least Squares (OLS), the model achieved a high coefficient of determination (R² = 0.855), a low Mean Absolute Error (MAE = 0.216), and Root Mean Squared Error (RMSE = 0.329). These results indicate a strong predictive performance based on limited but interpretable features. The findings reveal that narrative structure especially arc classification and viewer engagement contribute significantly to the perceived quality of episodes. While vote counts show limited correlation with ratings, combining them with narrative elements yields reliable predictions. This research offers a novel contribution to anime-based media analytics, emphasizing that minimal feature sets can provide robust predictive insight. The study also opens opportunities for enhancing content strategies and viewer understanding in serialized storytelling.

Predicting the Popularity Level of Roblox Games Using Gameplay and Metadata Features with Machine Learning Models

Ding Yi, Luo Jun, S Govindaraju — Tue, 01 Apr 2025 00:00:00 +0000

The online gaming platform Roblox has become a significant player in the gaming industry, providing a space for user-generated content. Predicting the popularity of Roblox games can help developers design better games and optimize user engagement. This study explores the use of machine learning models to predict the popularity of games on Roblox using gameplay features and metadata. A dataset of 9,734 games was collected, including variables such as likes, visits, game age, and active players. Three machine learning models, Decision Tree, Random Forest, and Gradient Boosting were employed to predict the number of favorites, which serves as a proxy for game popularity. Among the models tested, Gradient Boosting outperformed the others, achieving the highest R-squared score (0.85) and the lowest Root Mean Squared Error (11,470). Key features such as likes, game age, and visits were identified as the most influential in predicting game popularity. Based on these findings, this study recommends that developers focus on features that increase player engagement, such as regular updates and optimizing game exposure. Additionally, incorporating additional data sources, such as user reviews, and exploring explainability methods like SHAP can further improve model accuracy and transparency. This research contributes valuable insights into how machine learning can support decision-making in the development and optimization of Roblox games.

Anime Segmentation Based on User Preferences: Applying Clustering to Identify Groups of Anime with Similar Genres, Themes, and Popularity

Riswan E Tarigan, Yoana Sonia Wijaya — Tue, 01 Apr 2025 00:00:00 +0000

The anime industry has experienced significant growth, with an increasing focus on user preferences for content discovery and engagement. This study applies clustering techniques, specifically K-means, to segment anime based on user preferences, genres, themes, and popularity. By analyzing a comprehensive dataset containing attributes such as user ratings, popularity, genres, and themes, the research identifies distinct groups of anime that align with varying viewer tastes. The clustering results reveal that anime can be categorized into several groups, including highly popular but critically less-acclaimed titles, well-regarded but moderately popular anime, and niche, critically acclaimed series that appeal to smaller but dedicated audiences. This segmentation allows streaming platforms to offer more personalized recommendations, enhancing user experience and engagement by matching viewers with content that best fits their preferences. Although clustering techniques provide valuable insights into anime content, the study acknowledges certain limitations, such as overlap between clusters, indicating that some anime may not fit perfectly into a single category. This highlights the need for further improvements in segmentation accuracy. The study suggests exploring hybrid clustering methods, combining K-means with other techniques, and integrating demographic data, such as age, gender, and geographic location, to refine recommendations. Overall, the application of clustering algorithms to better understand user preferences in anime offers a promising approach to developing more effective and personalized recommendation systems. This can ultimately improve user satisfaction and engagement in the rapidly growing and competitive anime streaming market.

K-Means Clustering for Segmenting AI Survey Respondents: Analysis of Information Sources and Impact Perceptions

Evelyn, Satrio Pradono Suryodiningrat, Masmur Tarigan — Tue, 01 Apr 2025 00:00:00 +0000

This study employs K-Means clustering to analyze survey data from 91 university students, aiming to segment respondents based on their information-seeking behaviors (Question 2) and impact perceptions (Question 3) of artificial intelligence (AI). Two distinct clusters emerged: “Optimistic Problem Solvers,” who favor formal channels such as scholarly websites, peer-reviewed papers, and guided discussions, and express strong confidence in AI’s problem-solving capabilities with low concern for job displacement or dehumanization; and “Critical Watchers,” who rely more on informal, rapidly updated media (e.g., social platforms, general web searches) and exhibit heightened apprehension regarding AI’s socio-economic and ethical risks. Demographically, the former group skews toward sophomores with consistent GPAs and quantitatively oriented majors, while the latter displays broader disciplinary representation, balanced gender composition, and greater academic variability. These findings validate a dual-dimensional segmentation framework that integrates source behavior with perceptual orientation, highlighting the inadequacy of one-size-fits-all AI education. The study recommends differentiated instructional strategies, deep-dive, research-oriented modules for problem-solvers and trust-building, narrative-driven outreach for watchers, and outlines future research directions including larger, multi-institutional samples, longitudinal tracking, and mixed-methods approaches to refine and validate these profiles.