INTENSIF: Jurnal Ilmiah Penelitian dan Penerapan Teknologi Sistem Informasi
Not a member yet
174 research outputs found
Sort by
Implementation of SMOTE to Improve the Performance of Random Forest Classification in Credit Risk Assessment in Banking
Full text linkBackground: Credit is essential in banking operations, facilitating investment, corporate expansion, and financial satisfaction. Credit risk may emerge if the borrower defaults on payment commitments. Objective: This study aims to evaluate an individual\u27s creditworthiness by classifying and assessing their eligibility for credit. Methods: This study uses the Random Forest technique to categorize credit risk evaluation. Random Forest is a decision tree technique recognized for its high accuracy in data classification, utilizing an ensemble method of many decision trees. Before executing the classification process, issues frequently arise when data cannot be directly processed due to class imbalance. This study employs the SMOTE (Synthetic Minority Over-sampling Technique) algorithm to address class imbalance. The SMOTE algorithm is a method that emphasizes oversampling and is designed to augment the data in the minority class by generating synthetic data that aligns with the minority class data. The findings indicated that the ideal ratio for partitioning training and testing data was 80:20, and implementing the SMOTE technique within Random Forest enhanced performance assessment. Results: This research contributes to improving the accuracy of credit risk classification using the Random Forest algorithm, which effectively handles complex data and is supported by the implementation of SMOTE to overcome the class imbalance in the data. The classification accuracy value rose from 91.54% to 94.41%. The precision value rose from 90.83% to 97.03%, while the recall value increased from 60.26% to 91.55%. Conclusion: This method helps banks identify high-risk debtors more objectively and efficiently and supports appropriate credit decision-making
Development of Drip Irrigation Monitoring and Control System Model Based on the Internet of Things Using Android Applications
Full text linkBackground: Efficient water management is crucial for sustainable agriculture, particularly in regions with limited water resources. Drip irrigation systems, when integrated with the Internet of Things (IoT), offer a promising solution to optimize water usage and enhance agricultural productivity. Objective: This study aims to develop an IoT-based drip irrigation system to improve water efficiency and support small-scale farmers by providing a cost-effective and adaptable solution. Methods: The system employs multiple sensors to monitor key environmental parameters, including soil moisture, air temperature, and water levels in the tank. The collected data is transmitted to the ThingSpeak cloud platform via an Android application for real-time monitoring and control. Performance metrics such as sensor reaction time, solenoid valve response time, and pump response time were analyzed to evaluate system effectiveness. Results: Experimental findings show that the system effectively monitors and regulates irrigation, with an average sensor reaction time of 2.95 seconds, a solenoid valve response time of 2.75 seconds, and a pump response time of 2.3 seconds. The system successfully automates irrigation, ensuring optimal water usage. Conclusion: The IoT-based drip irrigation system enhances water resource management, increases crop yield, and reduces operational costs. While the system demonstrates high efficiency, further research could focus on scalability, integration with predictive analytics, and adaptation to different crop types and environmental conditions
Artificial Bee Colony-Based Optimization for Public Electric Vehicle Charging Station Placement
Full text linkBackground: The urgency of developing Electric Vehicle Charging Stations (EVCS) infrastructure is increasing alongside the need for low-emission mobility and energy efficiency. Objective: This study aims to optimize the determination of EVCS locations using the Artificial Bee Colony (ABC) method. Methods: This method was selected for its capability to find optimal solutions through an iterative population-based approach. Simulations were conducted by limiting the maximum iterations to 1000 to evaluate the impact of iteration numbers on optimization quality. Results: The results show that the ABC method successfully identified the shortest distance from three initial locations to the optimal EVCS locations. In the second simulation, the shortest distance obtained was 0.6420 km, indicating that an increase in the number of iterations correlates directly with the quality of optimization results. Specifically, the optimal distance from the first initial location to the EVCS at Danareksa Tower was 1.7018 km using the ultra-fast charging type. From the second initial location to the EVCS at the Ministry of State-Owned Enterprises Building, the optimal distance was 0.6420 km using the fast-charging type. Meanwhile, from the third initial location to the EVCS at PLN UID Greater Jakarta, the optimal distance was 1.1787 km using the ultra-fast charging type. Conclusion: This study demonstrates that the ABC method can deliver accurate results in determining optimal EVCS locations with efficient distances. These findings are expected to support the development of more effective and integrated electric vehicle infrastructure
Environmental Acoustic Features Robustness Analysis: A Multi-Aspecs Study
Full text linkAbstract—Background: Acoustic signals are complex, with temporal, spectral, and amplitude variations. Their non-stationarity complicates analysis, as traditional methods often fail to capture their richness. Environmental factors like reflections, refractions, and noise further distort signals. While advanced techniques such as adaptive filtering and deep learning exist, comprehensive acoustic feature analysis remains limited. Objective: This study investigates which acoustic features maintain the highest robustness across diverse environments while preserving discriminative power. Methods: Audio samples were recorded in controlled environments (jungles, cafés, factories, streets) with varying noise levels. Standardized equipment captured 22050 Hz, 16-bit audio at multiple positions and distances. After amplitude standardization, various acoustic features were extracted and analyzed. Results: MFCCs demonstrated exceptional reliability, with correlation coefficients of 0.98819 and 0.98889 for closely positioned devices and a robustness score of 0.99. Across different acoustic scenes and sample lengths (1, 3, 5s), MFCCs maintained high correlation (≈0.978) and robustness (0.98), confirming their versatility. Conclusion: MFCCs proved highly effective for acoustic fingerprinting across settings. Despite limitations in tested environments (≤5m distance, ≤5s samples), their consistent performance validates the methodology. Future research should explore combining MFCCs with spectral features and expanding studies to broader environments and device types
Sentiment Analysis of Suicide on X Using Support Vector Machine and Naive Bayes Classifier Algorithms
Full text linkBackground: The World Health Organization (WHO) defines health as a state of physical, mental, and social well-being, not just the absence of disease. Mental health, essential for overall well-being, is often neglected, leading to disorders like depression, a major cause of suicide. In Indonesia, suicide cases have surged, with 971 reported from January to October 2023. Objective: This study aims to analyze public sentiment regarding the rise in suicide cases in Indonesia using sentiment analysis methods, specifically Support Vector Machine (SVM) and Naive Bayes Classifier (NBC). The findings are expected to raise public awareness and provide policy recommendations to support mental health initiatives. Methods: One method used to understand public perception regarding the issue of suicide is text mining. This research employs text mining techniques with the Support Vector Machine (SVM) and Naive Bayes Classifier algorithms to analyze public sentiment related to suicide cases in Indonesia. Data was collected from tweets on social media platform X using crawling methods with snscrape and Python, totaling 1,175 tweets. Results: The results indicate that the Linear SVM model achieved higher accuracy than Naive Bayes in classifying tweet sentiments, with an accuracy rate of 80%. Conclusion: The SVM algorithm with a linear kernel achieved 80% accuracy and an identical ROC-AUC score. Word cloud visualization highlighted terms like "kill," "self," "depression," and "stress" as key negative sentiments. This study aims to raise public awareness and support better mental health policies in Indonesia
Machine Learning-Based Naïve Bayes Classification of Pineapple Productivity: A Case Study in North Sumatra
Full text linkBackground: Pineapple is a major agricultural commodity in Indonesia, especially in North Sumatra, where increasing demand calls for improved productivity. Although machine learning has been widely applied in agriculture, most prior studies on pineapple focus on fruit quality assessment or employ complex, less interpretable models, leaving a gap in lightweight and practical approaches for productivity classification. Objective: This study aims to evaluate the novelty and effectiveness of the Naïve Bayes algorithm in classifying pineapple productivity based on agronomic characteristics, addressing the underexplored use of this method for productivity prediction in pineapple cultivation. Methods: A descriptive quantitative approach was applied using secondary data from the Labuhan Batu Agricultural Extension Center, consisting of 52 records with seven agronomic parameters. The dataset was divided into 31 training and 21 testing samples, and the Naïve Bayes model was implemented using RapidMiner 7.1, with performance measured by accuracy. The small dataset size is recognized as a limitation that may affect generalizability. Results: The Naïve Bayes model achieved an accuracy of 86.67%, effectively distinguishing between productive and unproductive pineapples and demonstrating its suitability for agricultural classification tasks even with limited data. Conclusion: This study highlights the novelty and practicality of applying Naïve Bayes for pineapple productivity classification, offering an interpretable and computationally efficient alternative to more complex models. Future work should address dataset limitations by incorporating larger and more diverse samples and exploring hybrid or ensemble approaches to further enhance performance and support precision agriculture
Palm Oil Quality Based on Free Fatty Acid Using SVM
Full text linkBackground: Palm oil is one of the key commodities in both the food and non-food industries, with its quality largely influenced by the level of Free Fatty Acid (FFA). Obejctive: High FFA content can reduce the stability and market value of the oil. Classify palm oil quality based on FFA levels using the Support Vector Machine (SVM) algorithm. Methods: FFA levels were measured across multiple samples with varying usage frequencies (0, 5, 7, and 9 cycles) using the alkalimetric titration method. The measured data was categorized as "Suitable" if FFA ≤ 0.3% and "Unsuitable" if it exceeded this threshold. The developed SVM model was trained using 70% of the data and tested with the remaining 30%. Results: Evaluation results indicate that the model achieved an accuracy of 95%, a precision of 92%, and a recall of 94%, demonstrating SVM\u27s effectiveness in classifying data. Additionally, hyperplane visualization using PCA provided a clearer distinction between oil categories based on FFA levels. Conclusion: This study highlights that SVM can serve as an effective alternative for FFA-based palm oil quality classification. The implementation of this model is expected to enhance efficiency in the palm oil industry, particularly
Smart Governance Decision-Support System for Fisheries Development in Southeast Maluku: A Conceptual Framework
Full text linkBackground: Southeast Maluku Regency has vast marine and fishery resources; hence, the fisheries sector has not been a major economic contributor. The fisheries sector is still below its maximum capacity; this problem is caused by unsustainable fishing sector development planning. Objective: This research aimed to build framework tools to help plan and manage a sustainable and integrated fisheries sector based on empirical conditions. Methods: In this research, a suitable application framework was designed to support the development and planning of the fisheries sector in this region, the design of the input process, the input used, the interface, and the output produced to achieve smart government and a smart city. Results: This study built a conceptual framework tailored to the empirical conditions of the region in terms of geographical location and limited internet coverage for the Southeast Maluku Regency fisheries supporting master plan. Conclusion: The study provides guidance for researchers and practitioners in similar small island regions worldwide to construct a web-based intelligent DSS (decision support system) consistent with geographical conditions for planning the fisheries and marine sectors in their respective regions. The conceptual framework is adaptive which based on empirical condition both data and assessment of ranking for suitability location
Student Dropout Prediction Using Random Forest and XGBoost Method
Full text linkBackground: The increasing dropout rate in Indonesia poses significant challenges to the education system, particularly as students advance through higher education levels. Predicting student attrition accurately can help institutions implement timely interventions to improve retention. Objective: This study aims to evaluate the effectiveness of the Random Forest and XGBoost algorithms in predicting student attrition based on demographic, socioeconomic, and academic performance factors. Methods: A quantitative study was conducted using a dataset of 4,424 instances with 34 attributes, categorized into Dropout, Graduate, and Enrolled. The performance of Random Forest and XGBoost was compared based on accuracy, specificity, and sensitivity. Results: Random Forest achieved the highest accuracy at 80.56%, with a specificity of 76.41% and sensitivity of 72.42%, outperforming XGBoost. While XGBoost was slightly less accurate, it remained a competitive approach for student attrition prediction. Conclusion: The findings highlight Random Forest\u27s robustness in handling extensive datasets with diverse attributes, making it a reliable tool for identifying at-risk students. This study underscores the potential of machine learning in addressing educational challenges. Future research should explore advanced ensemble techniques, such as the Ensemble Voting Classifier, or deep learning models to further enhance prediction accuracy and scalability.
Identifying Key Features in Yelp Data for Success in Different Types of Restaurants
Full text linkBackground: The purpose of this research is to measure of customer satisfaction for newly established independent restaurants and, consequently, good predictors of independent restaurant success. Urban communities face several challenges, including how to best use scarce resources like real estate and support small enterprises. Smart businesses are essential to the development of smart cities because they use data analytics to inform their strategic planning and design choices, and the target of this topic is restaurant. Objective: Restaurants control a sizable portion of the city market\u27s small business sector. As part of the Yelp Data Challenge, Yelp just made available an open dataset that includes important details, ratings, and Yelp scores for every restaurant in different cities. Methods: Our methodology utilizes a vector of crucial factors to accurately forecast a business’s prospective success and exclusively evaluate eateries located inside the city limits of Las Vegas. The dependent variables will consist of the mean Yelp ratings for each restaurant and constructed our model by following the subsequent stages. Conclusion: The findings of this research is corroborated by the discovery that the statistically significant properties of restaurants, shown by a low p-value, varied across various restaurant categories, the unique modeling technique to forecast future restaurants\u27 Yelp rankings based on their design choices. This will assist owners of restaurants in making better design choices, which will result in more prosperous small enterprises in urban settings