Jurnal RESTI (Rekayasa Sistem dan Teknologi Informasi)
Not a member yet
1071 research outputs found
Sort by
An Investigation Towards Resampling Techniques and Classification Algorithms on CM1 NASA PROMISE Dataset for Software Defect Prediction
Get PDFSoftware defect prediction is a practical approach to improving the quality and efficiency of software testing processes. However, establishing robust and trustworthy models for software defect prediction is quite challenging due to the limitation of historical datasets that most developers are capable of collecting. The inherently imbalanced nature of most software defect datasets also posed another problem. Therefore, an insight into how to properly construct software defect prediction models on a small, yet imbalanced, dataset is required. The objective of this study is therefore to provide the required insight by way of investigating and comparing a number of resampling techniques, classification algorithms, and evaluation measurements (metrics) for building software defect prediction models on CM1 NASA PROMISE data as the representation of a small yet unbalanced dataset. This study is comparative descriptive research. It follows a positivist (quantitative) approach. Data were collected through observation towards experiments on four categories of resampling techniques (oversampling, under sampling, ensemble, and combine) combined with three categories of machine learning classification algorithms (traditional, ensemble, and neural network) to predict defective software modules on CM1 NASA PROMISE dataset. Training processes were carried out twice, each of which used the 5-fold cross-validation and the 70% training and 30% testing data splitting (holdout) method. Our result shows that the combined and oversampling techniques provide a positive effect on the performance of the models. In the context of classification models, ensemble-based algorithms, which extend the decision tree classification mechanism such as Random Forest and eXtreme Gradient Boosting, achieved sufficiently good performance for predicting defective software modules. Regarding the evaluation measurements, the combined and rank-based performance metrics yielded modest variance values, which is deemed suitable for evaluating the performance of the models in this context
Performance Analysis of MobileNetV3-based Convolutional Neural Network for Facial Skin Disorder Classification
Get PDFAccurately identifying facial skin types is essential for recommending the right skincare treatments and products. Misidentifying skin types can lead to negative consequences, such as irritation or worsening of skin conditions. This study investigated methods for classifying facial skin types into five categories: oily, acne-prone, dry, normal, and combination. A dataset of 1725 augmented facial images was used. Data augmentation techniques likely increased the dataset's diversity, which helps improve the model's generalization ability. The data underwent preprocessing, including rescaling, before being applied to two deep learning models, CNN and MobileNetV3. The models were evaluated based on accuracy and execution time to determine the most effective approach for classifying facial skin types. The CNN model achieved an accuracy of 64%, demonstrating its potential for image classification tasks. However, the MobileNetV3 model significantly outperformed CNN with an accuracy of 84%. This superior performance is attributed to MobileNetV3's advanced architecture, which is optimized for efficient feature extraction, and particularly relevant for capturing the subtle variations in facial skin types. Therefore, MobileNetV3 emerged as the more effective method for classifying facial skin types with higher accuracy
Optimizing Indonesian-Sundanese Bilingual Translation with Adam-Based Neural Machine Translation
Get PDFThis research seeks to construct an automatic translation between Indonesian and Sundanese languages based on the Neural Machine Translation (NMT) method. The model used in this study is the Long Short-Term Memory (LSTM) type, which carries out an encoder-decoder structure model learned with Bible data. The text translation here was conducted in different epochs to optimize the process, followed by the Adam optimization algorithm. Testing the Adam optimizer with different epoch settings yields a BLEU score for Indonesian to Sundanese translations of 0.991785, higher than the performance of the None optimizer. Experimental results demonstrate that Indonesian to Sundanese translation using Adam optimization with 1000 epochs consistently performed better in BLEU - Bilingual Evaluation Understudy - scoring than Sundanese to Indonesian translation. Limitations of the research were also put forth, particularly technical issues related to the collection of data and the Sundanese language’s complex grammatical features, that the model can only partially express, honorifics, and the problem of polysemy. Also, it must be mentioned that no special hyperparameter selection was performed, as parameters were chosen randomly. In future studies, transformer-based models can be investigated since these architectures will better deal with complex language via their self-attention mechanism
Comparison of Matrix Decomposition in Null Space-Based LDA Method
No full textProblems with small sample sizes and high dimensionality are common in pattern recognition. Almost all machine learning algorithms degrade in high-dimensional data, so that singularities in the scatter matrices, the main problem of the Linear Discriminant Analysis (LDA) technique, might result. A null space-based LDA (NLDA) has been conceived to address the singularity issue. NLDA aims to maximize the distance between classes in the null space of the within-class scatter matrix. In the first research, the NLDA method was performed by computing the eigenvalue decomposition and singular value decomposition (SVD). This research led to several new implementations of the NLDA method that use other matrix decompositions. The new implementations include NLDA using Cholesky decomposition and NLDA using QR decomposition. This paper compares the performance of three NLDA methods that use different matrix decompositions, namely, SVD, Cholesky decomposition, and QR decomposition. Two sets of data were used in the experiments that used three different NLDA algorithms. To determine the performance of the NLDA methods, the classification accuracy of the three methods was measured using the confusion matrix. The results show that the NLDA method using SVD has the best performance when compared to the other two methods, achieving a precision of 77.8% accuracy for the Colon dataset and a precision of 98.8% accuracy for the TKI-resistance dataset
Optimizing Book Stocktaking Process: Integration of Mobile Robot QRCode Commands with SLiMS
No full textThis study presents a novel approach to enhance efficiency and precision in library management through the utilization of QR code technology. Integration of a mobile robot equipped with a QR code reader into the stocktake process, interfaced with the SliMS framework via an accessible API, lays the groundwork for an automated book inventory management system. This groundbreaking system enables the generation of dynamic QR code commands, facilitating seamless adjustments to bookshelf layouts. The autonomous and accurate movement of the mobile robot significantly reduces the time required for recording, allowing library staff to allocate more time to value-added tasks. The implementation of this method entailed configuring the mobile robot to navigate library aisles, scan QR codes on book spines, and transmit inventory data to the SliMS system in real time. Research findings indicate a notable decrease in inventory processing time, accompanied by an improvement in accuracy resulting from the eradication of manual data entry errors. Specifically, the calculated efficiency gain of approximately 66.81% highlights the substantial benefits of integrating the mobile robot scan QR code process compared to manual methods. In conclusion, the deployment of this automated book inventory management system, driven by QR code technology, marks a positive shift in library management practices, enhancing the efficiency of the book inventory process and overall operational effectiveness
Remote Sensing Scene Classification using ConvNeXt-Tiny Model with Attention Mechanism and Label Smoothing
No full textRemote Sensing Scene Classification (RSSC) is the discrete categorization of remote sensing images into various classes of scene categories based on their image content. RSSC plays an important role in many fields, such as agriculture, land mapping, and identification of disaster-prone areas. Therefore, a reliable and accurate RSSC algorithm is required to ensure the accuracy of land identification. Many existing studies in recent years have used deep learning methods, especially CNN combined with attention modules to solve this problem. This study focuses on solving the RSSC problem by proposing a deep learning-based method (CNN) with the ConvNeXt-Tiny model integrated with Efficient Channel Attention Module (ECANet) and label smoothing regularization (LSR). The ConvNeXt-Tiny model shows that a persistent superior outperforms the ‘large’ model in convinced metrics. The ConvNeXt-Tiny model also has a huge advantage in high-precision positioning and higher classification accuracy and localization precision in a variety of complicated scenarios of remote sensing scene recognition. The experiments in this study also aim to prove that the integration of the attention module and LSR into the basic CNN network can improve precision, because the attention module can strengthen important features and weaken features that are less useful for classification. The experimental results proved that the integration of ECANet and LSR in the ConvNeXt-Tiny base network obtained a higher precision of 0.38% in the UC-Merced dataset, 0.7% in the AID, and 0.4% in the WHU-RS19 dataset than the ConvNeXt-Tiny model without ECANet and LSR. The ConvNeXt-Tiny model with ECANet integration and LSR obtained an accuracy of 99.00±0.41% in the UC-Merced dataset, 95.08±0.20% in AID, and 99.50±0.31% in the WHU-RS19 dataset
A Quantum Perceptron: A New Approach for Predicting Rice Prices at the Indonesian Wholesale Trade Level
No full textThe wholesale rice trade in Indonesia encounters various challenges in forecasting prices. These challenges are influenced by factors such as weather, government policies, global market conditions, and other economic variables. Accurate price predictions are crucial for informing government policy in a timely manner. This research introduces a new approach that utilizes the Quantum Perceptron algorithm to forecast rice prices. The algorithm, an innovative method in quantum computing, is expected to enhance the efficiency and effectiveness of price predictions. Although the research is still in the analytical stage, the use of Quantum Perceptron shows promise in dynamically addressing the complexity of market data and the variability of factors affecting rice prices. The method focuses on developing models that can leverage quantum computing to process information more effectively than classical methods. By harnessing the unique properties of quantum mechanics, such as superposition and entanglement, Quantum Perceptron can identify complex patterns and optimize predictions of future rice prices. The research describes the implementation of quantum algorithms in the context of the Indonesian rice wholesale market, including the technical challenges encountered and future development prospects. The research utilizes quantum computing along with the perceptron algorithm. The researchers focused on analyzing the quantum perceptron algorithm because of the limited availability of quantum computing devices. The findings of this research are confined to analysis. In order to advance this research, the author recommends that future studies employ quantum devices to achieve more accurate prediction
Comparing Correlation-Based Feature Selection and Symmetrical Uncertainty for Student Dropout Prediction
No full textPredicting student dropout is essential for universities dealing with high attrition rates. This study compares two feature selection (FS) methods—correlation-based feature selection (CFS) and symmetrical uncertainty (SU)—in educational data mining for dropout prediction. We evaluate these methods using three classification algorithms: decision tree (DT), support vector machine (SVM), and naive Bayes (NB). Results show that SU outperforms CFS overall, with SVM achieving the highest accuracy (98.16%) when combined with SU Moreover, this study identifies total credits in the fourth semester, cumulative GPA, gender, and student domicile as key predictors of student dropout. This study shows how using feature selection methods can improve the accuracy of predicting student dropout, helping educational institutions retain students better
Kmeans-SMOTE Integration for Handling Imbalance Data in Classifying Financial Distress Companies using SVM and Naïve Bayes
Get PDFImbalanced data presents significant challenges in machine learning, leading to biased classification outcomes that favor the majority class. This issue is especially pronounced in the classification of financial distress, where data imbalance is common due to the scarcity of such instances in real-world datasets. This study aims to mitigate data imbalance in financial distress companies using the Kmeans-SMOTE method by combining Kmeans clustering and the synthetic minority oversampling technique (SMOTE). Various classification approaches, including Nave Bayes and support vector machine (SVM), are implemented on a Kaggle financial distress data set to evaluate the effectiveness of Kmeans-SMOTE. Experimental results show that SVM outperforms Nave Bayes with impressive accuracy (99.1%), f1-score (99.1%), area under precision recall (AUPRC) (99.1%), and geometric mean (Gmean) (98.1%). On the basis of these results, Kmeans-SMOTE can balance the data effectively, leading to a quite significant improvement in performance
The Design of a C1 Document Data Extraction Application Using a Tesseract-Optical Character Recognition Engine
Get PDFThe 2019 election process used the Vote Counting Information System, also known as Sistem Informasi Penghitungan Suara (Situng), to provide transparency in the recapitulation process. The data displayed in Situng is from document C1 for 813,336 voting stations in Indonesia. The data collected from the C1 document is entered and uploaded into Situng by the officers of the Municipal General Election Commission (GEC). Since this process is performed by humans, it is not immune to errors. In the recapitulation process of the 2019 election results, there were 269 data entry errors, and the data entry process also did not run according to the specified target, resulting in delays. Furthermore, there were cases of C1 document modification, raising concerns about the data's authenticity. To avoid human errors and increase data entry speed, automatic data entry is a plausible option. The data entered are text data in image documents with the same template format, so that optical character recognition (OCR) can be used to read the text while improving image quality and alignment, resulting in a more accurate OCR reading area. In this study, we developed a C1 document data extraction application using the waterfall SDLC method, which has undergone a systematic and thorough process. The application was developed using Tesseract optical character recognition. Tesseract is an open-source OCR engine and command-line program that allows for the recognition of text characters within a digital image. The accuracy obtained by using this method is still not optimal as a substitute for Situng's data entry officer. To guarantee the integrity of the C1 document, we use the RSA-2048 digital signature scheme. The use of the Tesseract-OCR Engine for character recognition, combined with digital signature capabilities, provides a comprehensive solution to reduce the human error factor that can lead to miscalculations and inaccurate processes