Jurnal RESTI (Rekayasa Sistem dan Teknologi Informasi)
Not a member yet
1071 research outputs found
Sort by
Feature Selection Using Pearson Correlation for Ultra-Wideband Ranging Classification
No full textIndoor positioning plays a crucial role in various applications, including smart homes, healthcare, robotics, and asset tracking. However, achieving high positioning accuracy in indoor environments remains a significant challenge due to obstacles that introduce NLOS conditions and multipath effects. These conditions cause signal attenuation, reflection, and interference, leading to decreased localization precision. This research addresses these challenges by optimizing feature selection LOS, NLOS, and multipath classification within Ultra-Wideband (UWB) ranging systems. A systematic feature selection approach based on Pearson correlation is employed to identify the most relevant features from an open-source dataset, ensuring efficient classification while minimizing computational complexity. The selected features are used to train multiple machine-learning classifiers, including Random Forest, Ridge Classifier, Gradient Boosting, K-Nearest Neighbor, and Logistic Regression. Experimental results demonstrate that the proposed feature selection method significantly reduces model training and testing times without compromising accuracy. The Random Forest and Gradient Boosting models exhibit superior performance, maintaining classification accuracy above 90%. The reduction in computational overhead makes the proposed approach highly suitable for real-time applications, particularly in edge-computing environments where processing efficiency is critical. These findings highlight the effectiveness of Pearson correlation-based feature selection in improving UWB-based indoor positioning systems. The optimized feature set facilitates robust LOS, NLOS, and multipath classification while reducing resource consumption, making it a promising solution for scalable and real-time indoor localization applications
Optimizing Multilayer Perceptron for Car Purchase Prediction with GridSearch and Optuna
No full textMultilayer Perceptron (MLP) is a powerful machine learning algorithm capable of modeling complex, non-linear relationships, making it suitable for predicting car purchasing power. However, its performance depends on hyperparameter tuning and data quality. This study optimizes MLP performance using GridSearch and Optuna for hyperparameter tuning while addressing data imbalance with the Synthetic Minority Over-sampling Technique (SMOTE). The dataset comprises demographic and financial attributes influencing car purchasing power. Initially, the dataset exhibited class imbalance, which could lead to biased predictions; SMOTE was applied to generate synthetic samples, ensuring a balanced class distribution. Two hyperparameter tuning approaches were implemented: GridSearch, which systematically explores a predefined parameter grid, and Optuna, an adaptive optimization framework utilizing a Bayesian approach. The results show that Optuna achieved the highest accuracy of 95.00% using the Adam optimizer, whereas GridSearch obtained the best accuracy of 94.17% with the RMSProp optimizer, demonstrating Optuna's superior ability to identify optimal hyperparameters. Additionally, SMOTE significantly improved model stability and predictive performance by ensuring adequate class representation. These findings offer insights into best practices for optimizing MLP in predictive modeling. The combination of SMOTE and advanced hyperparameter tuning techniques is applicable to various domains requiring accurate predictive analytics, such as finance, healthcare, and marketing. Future research can explore alternative optimization algorithms and data augmentation techniques to further enhance model robustness and accuracy
Modified Particle Swarm Optimization on Feature Selection for Palm Leaf Disease Classification
No full textPalm oil plantations in Indonesia face challenges in enhancing productivity and profitability, notably due to pest attacks that reduce production. Early identification and classification of plant conditions, particularly palm oil leaves, are crucial for mitigating losses. This study explores the application of artificial intelligence, specifically computer vision and machine learning, for disease detection. Various machine learning techniques, including Local Binary Pattern (LBP), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM), have been used in different studies with varying accuracy. This research focuses on modifying Particle Swarm Optimization (PSO) for feature selection in identifying diseases in palm oil leaves. The PSO modification combined with logistic regression and Bayesian Information Criterion (BIC) significantly enhances KNN performance. Accuracy improved from 95.75% to 97.85%, while precision, recall, and F1-score reached approximately 98.80%. Additionally, the modified KNN+PSO achieved the shortest computation time of 0.0872 seconds, indicating high computational efficiency. These results demonstrate that the PSO modification not only improves accuracy but also computational efficiency, making it an effective method for enhancing KNN performance in detecting palm oil leaf diseases
Performance and Efficiency Comparison of U-Net and Ghost U-Net in Road Crack Segmentation with Floating Point and Quantization Optimization
No full textThis study presents a comprehensive comparison of U-Net and Ghost U-Net for road crack segmentation, emphasizing their performance and memory efficiency across various data representation formats, including FP32, FP16, and INT8 quantization. A dataset of 12,480 images was used, with preprocessing steps such as binarization and normalization to improve segmentation accuracy. Results show that Ghost U-Net achieved a marginally higher performance, with an IoU of 0.5041 and a Dice coefficient of 0.6664, compared to U-Net’s IoU of 0.5034 and Dice coefficient of 0.6662. Ghost U-Net also demonstrated significant memory efficiency, reducing GPU usage by up to 60% in FP16 and INT8 formats. However, a sharp decline in performance was observed for Ghost U-Net in the INT8 format, where the IoU dropped to 0.2038 and the Dice coefficient to 0.3227, whereas U-Net maintained stable performance across all formats. These findings suggest that Ghost U-Net is preferable for applications prioritizing memory efficiency and inference speed, while U-Net may be better suited for tasks requiring consistent accuracy across different quantization levels. This study underscores the importance of considering both performance stability and memory efficiency when selecting models for deployment in real-world applications
Max Depth Impact on Heart Disease Classification: Decision Tree and Random Forest
Get PDFResults in heart disease classification that are inaccurate and have low accuracy can endanger the patient's life. Some parameters in the algorithm model also influence classification. This study compares the Decision Tree and Random Forest algorithms for heart disease. The influence of maximum depth on heart disease classification also has significant implications. If the maximum depth is not set correctly, the classification results can be inaccurate and lead to incorrect diagnoses. This study uses five data split schemes, namely 60%: 40%, 70%: 30%, 75%: 25%, 80%: 20%, 90%: 10% and tested with different max depth parameters, namely max depth = 3, 4, 5, 6, and 7. This research produces the best accuracy using the 90%:10% scheme and max depth = 7 with the best accuracy result using the Random Forest algorithm of 99.29% while the Decision Tree algorithm is 98.05%. Then the precision and recall value of the Random Forest algorithm is 99% while the Decision Tree is 98%. The results of computation time using Decision Tree are faster than using Random Forest with a computation time for training data of 0.0075 s, while the testing data are 0.009 s. In future research, research can be conducted on the effect of other parameters by testing using several data sets
Comparison of Machine Learning Algorithms in Detecting Tea Leaf Diseases
Get PDFTea is one of the top ten export products sent from Indonesia to foreign countries. However, in recent years, the amount of tea leaf exports from Indonesia has decreased, although the value of the export impacts the country’s economic structure. In addition to market competition, Indonesia must maintain tea leaf production so that the increase in export decline is not significant or even increases tea leaf export production. To improve production quality and reduce production costs, early detection of tea leaf diseases is necessary. This study aims to classify tea leaf images for early detection of tea leaf disease so that appropriate treatment can be carried out early. This study compares machine learning algorithms to determine the best algorithm for detecting tea leaf diseases. The algorithms tested as performance comparisons in classifying tea leaf diseases are random forest (RF), support vector classifier (SVC), extra tree classifier (ETC), decision tree (DT), XGBoost classifier (XGB), and convolutional neural algorithms. Network (CNN). As a result, the average accuracy performance generated by ETC produces a higher value than other algorithms, i.e., getting an average accuracy performance of 77.47%. Another algorithm, SVC, has an average accuracy of 76.57%, RF of 76.12%, DT of 65.31%, XGB of 71.62%, and the lowest is CNN of 59.08%. ETC has been proven to be the most superior machine learning algorithm for detecting tea leaf diseases in this study
Comparative Analysis of Hybrid Model Performance Using Stacking and Blending Techniques for Student Drop Out Prediction In MOOC
Get PDFDespite being in high demand as a lifelong learner and academic material supplement, the implementation of Massive Open Online Courses (MOOC) has problems, one of which is the dropout rate (DO) of students, which reaches 93%. As one of the solutions to this problem, machine learning can be utilized as a risk management and early warning system for students who have the potential to drop out. The use of ensemble techniques to build models can improve performance, but previous research has not reviewed the most optimal ensemble technique for this case study. As a form of contribution, this study will compare the performance of models built from stacking and blending techniques. The algorithms used in the base model are KNN, Decision Tree, and Naïve Bayes, while the meta-model uses XGBoost. These algorithms are used to build models with stacking and mixing techniques. The experimental results using stacking are 82.53% accuracy, 84.48% precision, 94.12% recall, and 89.04% F1 score. Meanwhile, the blend obtained 83.39% precision, 85.31% precision, 94.21% recall, and 89.54% F1-Score. These results are supported by model testing using k-fold cross-validation and confusion matrix techniques, which show the same results. That is, blending is 0.86% higher than stacking, so it can be concluded that blending performs better than stacking in the MOOC student dropout prediction case study
Twitter Sentiment Analysis Towards Candidates of the 2024 Indonesian Presidential Election
No full textLong before the elections were held, the topic related to elections was widely discussed on news portals and social media, including Twitter. A few studies related to the Indonesian election have tried to predict candidates who will run for the presidential election, but there has been no research that examines public sentiment on social media towards each of the potential candidates. The main objective of this study is to analyze the public sentiment in Twitter towards potential candidates for the 2024 Indonesian presidential election. This research seeks to fill the gaps in previous research and become a reference for further research regarding sentiment analysis for election prediction using Twitter. The presidential candidates used in the research are the top 3 candidates based on the Poltracking survey, namely Ganjar Pranowo, Prabowo Subianto, and Anies Baswedan. The data were taken from January until October 2022, more than a year before the general election began. To predict the sentiment, four different machine-learning methods were used and compared to each other. There are Naïve Bayes, Support Vector Machines, Random Forests, and Neural Networks. Based on the sentiment results of each candidate, the highest sentiment towards Prabowo is neutral (55.49%), the highest sentiment towards Ganjar is positive (61.34%), and the highest sentiment towards Anies is neutral (44.84%). Results from the study also show that Anies was the presidential candidate who received more negative sentiment than the other two (56.63%). Meanwhile, Ganjar Pranowo got the most positive sentiment of all (42,69%). For the neutral sentiment, Anies Baswedan also got the most results (39,87%), followed by Prabowo (38.99%) and Ganjar Pranowo (21.14%). The result of the study also discovered that random forest and neural networks have the best performance for sentiment analysis. 
Cloud Node Auto-Scaling System Automation Based on Computing Workload Prediction
No full textAuto-scaling systems in cloud computing are important for handling application workload fluctuations. This research uses machine learning to predict resource requirements based on workload work patterns and design an automatic scaling system. The dataset used includes features of node name, time, CPU usage percentage, and RAM usage. The ML model is applied for prediction regression of CPU usage percentage, CPU load, and RAM usage, and then server workload is classified into four categories: Very High, High, Low, and Very Low. The autoscaling system used is horizontal scaling. From the results of this research, it was found that the stacking algorithm with the base learner Random Forest and XGBoost had better performance in producing predictive regression. Then, after performing stability testing using K-Fold cross-validation by classifying based on workload status, it was found that the Gradient Boosting algorithm had better results compared to other algorithms, namely for the percentage of CPU usage with an accuracy of 0.998, precision 0.9, recall 0.878, f1score 0.888; CPU load average 15 minutes with accuracy 0.997, precision 0.854, recall 0.863, f1score 0.863; Meanwhile, the percentage of RAM usage is accuracy 0.992, precision 0.986, recall 0.986, and f1score 0.986. However, the XGBoost algorithm also has test results that are almost the same as Gradient Boosting
Manhattan Distance-based K-Medoids Clustering Improvement for Diagnosing Diabetic Disease
Get PDFDiabetes is a metabolic disorder characterized by blood glucose levels above normal limits. Diabetes occurs when the body is unable to produce sufficient insulin to regulate blood sugar levels. As a result, blood sugar management becomes impaired and there is no cure for diabetes. Early detection of diabetes provides an opportunity to delay or prevent its progression into acute stages. Clustering can help identify patterns and groups of diabetes symptoms by analyzing attributes that indicate these symptoms. In this study, researchers are using K-Medoid and Quantum K-Medoid methods for clustering diabetes data. Quantum computing utilizes quantum bits, or qubits, which can represent multiple states at the same time. Compared to classical computers, quantum computing has the potential for an exponential speedup in problem-solving. Researchers conducted a comparison between two methods: the classic K-Medoids method and the K-Medoids method utilizing quantum computing. The researchers found that both Quantum K-Medoid and Classic K-Medoid achieved the same clustering accuracy of 91%. In testing with the Quantum K-Medoids algorithm, it was found that the cost value in the 8th epoch showed a significant decrease compared to the Classical K-Medoids algorithm. This demonstrates that Quantum K-Medoid can be considered a viable alternative for clustering purposes