Jurnal Aplikasi Sains Data

Analysis of the Impact of Gadget Use on Student Academic Productivity

2026-10-23T16:14:20+07:00

The increasing use of gadgets among students provides high flexibility in supporting academic activities, but on the other hand, it has the potential to trigger distractions that have an impact on reducing learning productivity. This research aims to analyze the influence of gadget usage patterns on the duration of student learning productivity using a quantitative approach. Data collection was carried out using a Likert scale questionnaire, which was distributed to students from various faculties at the National Development University "Veteran" East Java. Data analysis included multiple linear regression to test the influence of quantitative variables and attitude scales simultaneously or partially, as well as the One-Way ANOVA test to evaluate differences in average productivity based on groups of categorical variables. Before modeling was carried out, all data had gone through the cleaning stages and testing of classical prerequisite assumptions, which included tests for normality, multicollinearity, heteroscedasticity, and autocorrelation. The results of the regression analysis show that together the variables of perceived duration of task work, level of lost time, and ability to ignore social media have a significant effect on the duration of learning productivity. Partially, the perception of lost time due to the use of irrelevant devices is the main obstacle to productivity, while the ability to limit oneself from social media and focus on completing tasks makes a real positive contribution. On the other hand, the results of the One-Way ANOVA test prove that there is no significant difference in the average duration of learning productivity, both in terms of variations in the time when distractions occur and the main reason why students open their devices. These findings show that the impact of technology is universal, where student learning productivity is determined more by internal self-control and time management than conditional factors such as gadget use.

A Comparative Analysis of K-Medoids and DBSCAN for Clustering Student Consumptive Behavior Profiles

2026-10-22T14:30:35+07:00

The rapid adoption of digital payment technologies, such as e-wallets and mobile banking, has significantly altered the financial habits of university students, often blurring the psychological boundaries of spending compared to conventional cash transactions. This study aims to objectively map and cluster the multidimensional consumptive behavior profiles of university students based on their payment preferences and living status. A quantitative approach was employed using a dataset of 150 respondents evaluated across 15 financial and behavioral variables. To address the complexity of the high-dimensional data, preprocessing techniques including One-Hot Encoding, Min-Max Scaling, and Principal Component Analysis (PCA) were applied prior to modeling. The study utilized a comparative clustering approach to evaluate the performance of the K-Medoids and DBSCAN algorithms. The evaluation metrics demonstrate that DBSCAN (, MinPts ) outperformed K-Medoids, achieving an optimal Silhouette Score of 0.5438, a Davies-Bouldin Index of 0.7151, and a Calinski-Harabasz Index of 101.57. The DBSCAN model successfully mapped the data into six distinct consumer clusters. The profiling results reveal a significant behavioral divide: students preferring digital payments exhibit notably higher consumptive tendencies, impulsivity, and reliance on PayLater credit facilities compared to highly conservative cash users. Furthermore, living status dictates fixed expense allocations, compounding the financial vulnerabilities of specific groups. These empirical findings provide a robust data-driven foundation for developing targeted financial literacy strategies to help students manage digital financial ecosystems responsibly

Keywords: Consumptive Behavior, Clustering, DBSCAN, K-Medoids, Digital Payment.

Analysis of Preference Clustering and Strategies for Utilizing Sources and Media in Accessing Academic References by College Students Using the K-Means and DBSCAN Algorithms

2026-10-22T14:01:07+07:00

The increasing availability of academic information sources has led to diverse preferences and strategies among university students in searching for and utilizing academic references. This study aims to identify patterns in students’ academic reference-seeking behavior by applying Principal Component Analysis (PCA) and comparing the clustering performance of K-Means and Density-Based Spatial Clustering of Applications with Noise (DBSCAN). Data were collected through an online questionnaire involving 150 university students, with ten numerical variables used in the analysis. The data were preprocessed, standardized using StandardScaler, and reduced using PCA before clustering. The results showed that PCA reduced the ten variables to six principal components while retaining 83.53% of the cumulative variance. K-Means produced two clusters containing 123 and 27 respondents, achieving a Silhouette Score of 0.4543, a Calinski-Harabasz Index (CHI) of 52.2985, and a Davies-Bouldin Index (DBI) of 1.4225. Meanwhile, DBSCAN, with eps = 2.3 and min_samples = 2, identified two clusters consisting of 137 and 2 respondents, along with 11 noise points. DBSCAN achieved a Silhouette Score of 0.5469, a CHI of 15.3191, and a DBI of 0.4818. Therefore, DBSCAN was selected for further analysis to identify differences in students’ academic reference-seeking behaviors, particularly in source verification, reference usage, and manual and AI-assisted searching.

Applying Principal Component Regression to Factors Associated with College Students’ Frequency ofCooking Their Own Breakfast

2026-10-23T16:15:30+07:00

Cooking breakfast independently is associated with healthier eating habits, but its frequency among college. Several

interrelated factors influence students. This study aimed to model the weekly frequency of cooking breakfast independently

(Y) using principal component regression (PCR) with four predictors: confidence in cooking ability (X1), frequency of not

cooking because food ingredients had run out (X2), frequency of cooking because food outside the home was more expensive

(X3), and frequency of preparing breakfast to take as a packed meal (X4). The four predictors were selected from 15

candidate questions based on conceptual relevance and examined using Spearman’s correlation across 12 numerical

indicators; X1 and X3 were significantly correlated with Y, while X2 and X4 were retained despite not being significant.

The data came from 150 college students who completed an online form, with no missing values. The analysis included

descriptive statistics, Pearson’s correlation, standardisation, principal component analysis (PCA), and PCR, with the number

of components selected through 5-fold cross-validation based on the minimum RMSE. Two principal components explained

61.01% of the variance in the predictors and produced a cross-validation RMSE of 1.5882. The model was jointly significant

(F = 18.65; p = 6.03×10−8), with an R² of 0.2024, an adjusted R² of 0.1916, and a cross-validated R² of 0.1556. On the

original scale, X1, X3, and X4 were positively associated with Y, whereas X2 was negatively associated with Y. The model’s

explanatory power was low to moderate; therefore, interpret the results as associations rather than causal relationships.

Analysis of Campus Facility Availability and Academic Community Satisfaction Using Logistic Regression and Random Forest

2026-10-23T16:17:30+07:00

Campus facilities play an important role in supporting academic activities and student satisfaction. This study aimed to classify student satisfaction with campus facilities using Binary Logistic Regression and Random Forest. Survey data from 150 student respondents were used, with overall satisfaction scores of 1–3 classified as lower satisfaction and scores of 4–5 as higher satisfaction. Five facility-availability predictors were selected using Pearson correlation with a threshold of (|r| 0.20). The dataset was divided into training and testing sets using an 80:20 stratified split. Class weighting, five-fold stratified cross-validation, and GridSearchCV were applied to address class imbalance and optimize model hyperparameters. On the testing set, the tuned Random Forest achieved 93.33% accuracy, 83.33% precision, 83.33% recall, an F1-score of 83.33%, and a ROC-AUC of 86.81%. Logistic Regression achieved 90.00% accuracy, 71.43% precision, 83.33% recall, an F1-score of 76.92%, and a ROC-AUC of 92.36%. Random Forest was selected as the final model because it achieved higher test-set accuracy, precision, and F1-score than Logistic Regression. Feature importance analysis identified parking availability as the most important predictor, with an importance value of 0.3899. These findings suggest that Random Forest can classify student satisfaction in this dataset, although further validation using larger samples is needed to assess model generalizability.

Analysis of Music-Listening Habits and Mood Changes During Daily Activities Using Multiple Linear Regression

2026-10-23T16:19:37+07:00

Music is commonly used during daily activities for entertainment, concentration, and mood regulation. However, the relationship between music-listening habits and perceived mood changes may vary across individuals and listening contexts. This study aims to explore music-listening patterns and examine the association between selected listening habits and perceived mood changes using Exploratory Data Analysis (EDA) and multiple linear regression. The study used survey data from 150 respondents. Data preprocessing included removing personal identification columns, checking missing values and duplicate records, inspecting potential outliers using the Interquartile Range (IQR) method, and encoding categorical variables using one-hot encoding. EDA was conducted using descriptive statistics, distribution plots, scatter plots, and Pearson correlation analysis. The initial regression model included 11 predictors and produced an R² of 0.235 and an adjusted R² of 0.174, with an overall F-statistic of 3.863 and a p-value of 7.19 × 10⁻⁵. Diagnostic testing indicated non-normal residuals and heteroskedasticity. Therefore, the final model used heteroskedasticity-consistent HC3 standard errors and included two predictors: the percentage of daily activities perceived as more comfortable with music and the indicator for listening without a specific time. The final model achieved an R² of 0.1503 and an adjusted R² of 0.1387. Both predictors were statistically significant in the final model. These findings suggest that perceived comfort when listening to music and listening-time patterns are associated with reported mood changes. Nevertheless, the observational survey design and limited explanatory power mean that the results should not be interpreted as evidence of causation.

Keywords: exploratory data analysis, music-listening habits, mood changes, multiple linear regression, HC3 robust standard errors

Analysis of Study Space Comfort Levels in Coffee Shops Using Decision Tree and Random Forest Algorithms

2026-10-23T16:20:57+07:00

Coffee shops are no longer just places to enjoy drinks, they have also become workspaces and study spaces for university students. This study examines this shift through Exploratory Data Analysis (EDA) of questionnaire data from 150 valid respondents, covering 22 variables such as visit frequency and duration, spending, number of devices, seating area, preferred time, noise disturbance, Wi-Fi and power outlet problems, and the percentage of tasks completed. The data were cleaned and checked for missing values and outliers using the IQR method. The results show that most respondents feel comfortable, with scores of 4 and 5 being the most common, particularly among students from Computer Science and Engineering and Science, in the late afternoon, and in workspace areas. Comfort is positively related to productivity, with the percentage of tasks completed rising from 52.0% for Uncomfortable respondents to 64.9% for Comfortable respondents and 74.1% for Very Comfortable respondents. However, Wi-Fi and power outlet problems remain common, experienced occasionally by 41.3% and often by 33.3% of respondents. Decision Tree and Random Forest models reached 50.0% and 56.7% accuracy respectively in predicting comfort, so they are not yet reliable without further tuning. Overall, the findings confirm that coffee shops now serve as student workspaces, and facility quality is a factor owners should address.

Comparison of Support Vector Machine (SVM) and K-Nearest Neighbors (KNN) Methods in Classifying University Students’ Sleep Quality Based on Academic Assignment Workload

2026-10-23T15:48:31+07:00

A high academic assignment workload often directly reduces the duration and quality of students’ sleep, triggering sleep debt and a decline in both physical and psychological performance. This study aims to analyze the effect of academic assignment volume on students’ sleep duration and to classify their sleep quality using machine learning algorithms. Data were collected through a questionnaire from 150 respondents with 15 variables covering assignment work duration, sleep hours, and assignment pile-up factors. Exploratory Data Analysis (EDA) showed a negative correlation of -0.43 between assignment work hours and sleep duration on active lecture days. Classification modeling was carried out using the Support Vector Machine (SVM) algorithm with an optimized Radial Basis Function (RBF) kernel (C=100, γ=0.01) and K-Nearest Neighbors (KNN) with the best number of neighbors (k=25). The test results show that the SVM model was slightly better overall, with an accuracy of 56.67% and a macro F1-score of 56.62%, compared with the KNN model, which achieved an accuracy of 53.33% and a macro F1-score of 51.39%. In addition, SVM proved better at detecting the poor sleep class, while KNN was more sensitive to the restful sleep class. This study demonstrates that time management and the amount of assignment workload affect students’ rest duration, and that a machine learning approach is able to map these classification patterns despite the challenge of overlapping data.

ANALYSIS OF THE EFFECTS OF DISCOUNTS AND PROMOTIONS ON IMPULSE BUYING

2026-10-23T16:23:13+07:00

This study aims to analyze the effects of discount and promotional strategies on impulse buying tendencies. The development of consumer behavior, particularly on digital platforms such as Shopee, has made impulse buying an increasingly relevant phenomenon to investigate, especially in relation to price incentives offered by sellers. To achieve this objective, this study employs a quantitative approach by analyzing data from 150 respondents. The analytical methods include Principal Component Analysis (PCA) to reduce ten independent variables into a simpler set of components and Principal Component Regression (PCR) to examine the effects of these components on impulse buying tendencies.

The results indicate that PCA successfully reduced the dimensionality of the data. The first principal component (PC1) explained 33.49% of the total variance, while the first two components (PC1 and PC2) jointly explained 49.01%. By including the sixth component (PC6), the cumulative explained variance reached 83.48%, indicating that most of the information in the data could be represented by six principal components. In the model diagnostics, 10 out of 150 respondents had Cook's Distance values exceeding the established threshold, indicating the presence of several influential observations that should be considered when interpreting the stability of the regression model. Based on these findings, discount and promotional strategies were found to have a statistically significant relationship with impulse buying tendencies, although the presence of influential observations should be taken into account when interpreting the model results.

Keywords: Principal Component Analysis, Principal Component Regression, Diskon, Promo, E-Commerce

A Comparative Analysis of K-Nearest Neighbors and Logistic Regression for Classifying the Effectiveness of the Do Not Disturb Feature in Supporting Student Focus and Productivity

2026-10-23T16:30:38+07:00

Smartphone notifications can interrupt students’ concentration during academic activities, making the effectiveness of the Do Not Disturb (DND) feature relevant to focus and productivity. This study compares K-Nearest Neighbors (KNN) and Logistic Regression to classify perceived DND effectiveness using eleven predictors related to digital device use, study focus, DND usage, and perceived productivity. Data were collected through an online questionnaire involving 150 students at Universitas Pembangunan Nasional “Veteran” Jawa Timur. The target variable was grouped into effective and ineffective categories. The dataset was divided into training and testing sets using a stratified 70:30 split, and five-fold cross-validation was applied during model development. KNN achieved a cross-validation accuracy of 89.52% with k = 17, while Logistic Regression obtained 82.86%. On the 45-observation test set, Logistic Regression achieved an accuracy of 84.44%, precision of 88.24%, recall of 90.91%, F1-score of 89.55%, and ROC-AUC of 0.8990. KNN achieved 80.00% accuracy, 81.58% precision, 93.94% recall, 87.32% F1-score, and 0.9318 ROC-AUC. Thus, Logistic Regression performed better in accuracy, precision, and F1-score, whereas KNN had higher recall for the effective class and a higher ROC-AUC. Logistic Regression coefficients also showed that perceived DND assistance in maintaining focus was positively associated with the predicted effective category. Overall, Logistic Regression provided more balanced classification performance on the test set, although neither algorithm outperformed the other across all metrics. These findings concern students’ reported perceptions and do not establish a causal effect of DND on focus or productivity.

NEXT-DAY RAINFALL CATEGORY PREDICTION USING ENSEMBLE MACHINE LEARNING AND SHAP

2026-10-23T16:28:09+07:00

Rainfall intensity varies substantially over time and is difficult to classify using simple linear approaches because meteorological variables interact nonlinearly. This study develops a day-ahead rainfall-category classification framework using meteorological observations from BMKG for 2025. The dataset contains 366 daily observations of rainfall, temperature, humidity, wind speed, and wind direction. Temporal feature engineering was applied through three-, five-, and seven-day rainfall accumulation as well as rolling humidity and wind-speed features. Random Forest, XGBoost, and LightGBM were trained and compared using accuracy, weighted precision, weighted recall, and weighted F1-score. Hyperparameter tuning was performed using cross-validation, and SHAP was employed to interpret the selected model. XGBoost produced the best balance between precision and recall, with accuracy of 0.8356, weighted precision of 0.7710, weighted recall of 0.8356, and weighted F1-score of 0.8010. The five-fold cross-validation score was 0.8528. SHAP analysis identified humidity as the most influential feature, followed by seven-day rainfall accumulation and five-day rainfall accumulation. These findings indicate that short-term meteorological history and atmospheric humidity are important signals for day-ahead rainfall-category classification.