Jurnal RESTI (Rekayasa Sistem dan Teknologi Informasi)
Not a member yet
1071 research outputs found
Sort by
Investigating the Impact of ReLU and Sigmoid Activation Functions on Animal Classification Using CNN Models
Get PDFVGG16 is a convolutional neural network model used for image recognition. It is unique in that it only has 16 weighted layers, rather than relying on a large number of hyperparameters. It is considered one of the best vision model architectures. However, several things need to be improved to increase the accuracy of image recognition. In this context, this work proposes and investigates two ensemble CNNs using transfer learning and compares them with state-of-the-art CNN architectures. This study compares the performance of (rectified linear unit) ReLU and sigmoid activation functions on CNN models for animal classification. To choose which model to use, we tested two state-of-the-art CNN architectures: the default VGG16 with the proposed method VGG16. A dataset consisting of 2,000 images of five different animals was used. The results show that ReLU achieves a higher classification accuracy than sigmoid. The model with ReLU in fully connected and convolutional layers achieved the highest precision of 97.56% in the test dataset. The research aims to find better activation functions and identify factors that influence model performance. The dataset consists of animal images collected from Kaggle, including cats, cows, elephants, horses, and sheep. It is divided into training sets and test sets (ratio 80:20). The CNN model has two convolution layers and two fully connected layers. ReLU and sigmoid activation functions with different learning rates are used. Evaluation metrics include accuracy, precision, recall, F1 score, and test cost. ReLU outperforms sigmoid in accuracy, precision, recall, and F1 score. This study emphasizes the importance of choosing the right activation function for better classification accuracy. ReLU is identified as effective in solving the vanish-gradient problem. These findings can guide future research to improve CNN models in animal classification
Image Preprocessing Approaches Toward Better Learning Performance with CNN
No full textConvolutional neural networks (CNNs) are at the forefront of computer vision, relying heavily on the quality of input data determined by the preprocessing method. An undue preprocessing approach will result in poor learning performance. This study critically examines the impact of advanced image pre-processing techniques on computational neural networks (CNNs) in facial recognition. Emphasizing the importance of data quality, we explore various pre-processing approaches, including noise reduction, histogram equalization, and image hashing. Our methodology involves feature visualization to improve facial feature discernment, training convergence analysis, and real-time model testing. The results demonstrate significant improvements in model performance with the preprocessed dataset: average accuracy, recall, precision, and F1 score enhancements of 4.17%, 3.45%, 3.45%, and 3.81%, respectively. Additionally, real-time testing shows a 21% performance increase and a 1.41% reduction in computing time. This study not only underscores the effectiveness of preprocessing in boosting CNN capabilities, but also opens avenues for future research in applying these methods to diverse image types and exploring various CNN architectures for comprehensive understanding
Evaluating the Bibit App: The HEART Framework Approach in UX Design
No full textWith the rise of investment applications like Bibit in Indonesia, evaluating user experience is crucial to improve user engagement and satisfaction. This study aims to assess the user experience of the Bibit app using the HEART framework to identify areas for improvement. A descriptive survey was conducted using the Slovin method to determine a sample size of approximately 100 people. The study used a questionnaire distributed through Google Forms that measures the five aspects of the HEART framework on a Likert scale. Data were analyzed for validity and reliability using IBM SPSS and stored in Google Spreadsheets. In general, the Bibit app demonstrated high satisfaction, engagement, adoption, retention, and task success among its users, with most categories scoring above the 80% target. However, specific areas, particularly Happiness 2, Adoption 1, Retention 1, and Task Success 3, require further development attention. The Bibit app provides a generally satisfying user experience, with significant potential for improvement in navigation tooltips, investment product development, and app speed and efficiency
Reducing Training Time in Skin Cancer Classification Using Convolutional Neural Network with Mixed Precision Implementation
No full textIn the field of skin cancer classification, machine learning and deep learning have been extensively utilized, particularly with convolutional neural network (CNN) architectures. However, there remains room for exploration to achieve optimal performance. This study investigates the use of the MobileNetV3Large architecture for transfer learning, chosen for its efficiency in low-power and memory-constrained applications. To further enhance performance, black-hat morphological transformation and oversampling techniques were applied to the ISIC 2020 dataset. Additionally, mixed precision training was implemented to reduce training time. The research aimed to compare the accuracy, precision, recall, F1-score, and training time of models trained with and without mixed precision. The findings revealed that while the model without mixed precision achieved superior performance with accuracy, precision, recall, and F1-score metrics reaching 98%, both models yielded an AUC-ROC of 1. Notably, mixed precision training significantly reduced training time by 1,646 seconds (27 minutes and 26 seconds), representing an 8.39% speed increase. These results suggest that mixed precision can meaningfully accelerate model training while maintaining competitive performance. The practical implications of this research include its potential to improve the efficiency of skin cancer classification models, making them more suitable for real-time clinical applications, particularly in resource-constrained environments
Data Clustering for Sentiment Classification with Naïve Bayes and Support Vector Machine
No full textVisitor reviews play a crucial role in determining the success of a business, particularly those offering hospitality and services, such as hotels. The growth of internet technology has made it easier for guests to share their experiences, which can influence potential customers. Google Maps is one of the platforms used for giving and searching reviews This research uses data crawled from Google Maps Review using the playwright library. However, the large volume of reviews can make analysis and topic-based categorization—such as service quality, hotel location, and operational hours—challenging. To address this, DBSCAN is used to cluster reviews based on these topics. Clustering helps improve sentiment classification, making it more targeted and allowing a comparison of two machine learning algorithms: Naïve Bayes and Support Vector Machine (SVM). Naïve Bayes achieved higher accuracy (0.87) in the operational hours cluster, while SVM scored 0.78. However, SVM showed improved accuracy in the location (0.89) and service (0.88) clusters, with Naïve Bayes maintaining a stable 0.86 accuracy in both. Both models demonstrated an average training time of less than one second, excluding preprocessing
GLCM-Based Feature Extraction for Alpha Matting on Natural Images
No full textThe main objective of this research is to determine the optimal threshold value in the unknown region in the alpha-matting operation of natural images. Alpha-mating serves to draw matte from the image used in segmentation. The alpha value is very influential on the quality of segmentation which is determined by the level of threshold value accuracy. The determination of the threshold begins by breaking the grayscale image into several sub-images using Region of Interest (RoI). Each sub-image was extracted using the Gray Level Co-occurrence Matrix (GLCM) considered by the parameters of contrast, energy, and entropy at angles of 0°, 45°, 90°, and 135 °. Each feature results in extractions, which are then averaged and normalized in each sub-image. The value is determined as the local threshold value used in the alpha matting operation. Experiments were carried out on 12 natural images from the image-mating dataset to evaluate the performance of the proposed algorithm. The increase in accuracy shows up to 63% by the measurements of experiments, compared to the calculation of adaptive threshold by using the fuzzy CMs Algorithm
Quantum-Enhanced K-Medoids Clustering: Comparative Analysis of Stroke Medical Data
No full textStroke is a severe medical condition that occurs when the blood supply to parts of the brain is interrupted or reduced, resulting in brain tissue that lacks oxygen and nutrients. This causes brain cells to start to die in minutes. Early prevention reduces the risk of stroke. In this study, a quantum computing approach is used to improve the performance of the K-Medoids method. A comparative analysis of these methods was carried out with a focus on their performance, especially on the accuracy of the test results. The investigation was carried out using a data set of stroke patient medical records. The data set was tested using the classical and K-Medoids methods with a quantum computing approach utilizing Manhattan distance calculations. The findings of this research reveal improvements in the K-Medoids algorithm with Manhattan distance calculation influenced by the integration of a quantum computing framework. In particular, the simulation test results show an increase in accuracy from the classical K-Medoids method to the K-Medoids method with a quantum computing approach, from 52% to 64%. These results highlight that the performance of the K-Medoids method with a quantum computing approach is superior to that of the classical K-Medoids method
An Optimized Hyperparameter Tuning for Improved Hate Speech Detection with Multilayer Perceptron
No full textHate speech classification is a critical task in the domain of natural language processing, aiming to mitigate the negative impacts of harmful content on digital platforms. This study explores the application of a Multilayer Perceptron (MLP) model for hate speech classification, utilizing Bag of Words (BoW) for feature extraction. The hypothesis posits that hyperparameter tuning through sophisticated optimization techniques will significantly improve model performance. To validate this hypothesis, we employed two distinct hyperparameter tuning approaches: Random Search and Optuna. Random Search provides a straightforward yet effective means of exploring the hyperparameter space, while Optuna offers a more sophisticated, optimization-based approach to hyperparameter selection. The study involved training the MLP model on a labeled dataset is based on crawling results on the Twitter platform of hate speech and non-hate speech overall total dataset is 13.169, followed by evaluation using standard metrics. Our experimental results demonstrate the comparative effectiveness of these two hyperparameter tuning methods. Notably, the MLP model tuned with Optuna achieved a higher F1-score of 81.49%, compared to 79.70% achieved with Random Search, indicating the superior performance of Optuna in optimizing the hyperparameters. These results were obtained through extensive cross-validation to ensure robustness and generalizability. The findings underscore the importance of optimized hyperparameters in developing robust hate speech classification systems. The superior perform ance of Optuna highlights its potential for broader application in other machine learning tasks requiring hyperparameter optimization. This improvement enables more reliable and efficient automated moderation, which is crucial for the integrity and security of digital communication platforms such as Twitter
Data Mining Techniques for Predictive Classification of Anemia Disease Subtypes
No full textAnemia, characterized by insufficient red blood cells or reduced hemoglobin, hinders oxygen transport in the body. Understanding the various types of anemia is vital to tailor effective prevention and treatment. This research explores data mining's role in predicting and classifying anemia types, emphasizing Complete Blood Count (CBC) and demographic data. Data mining is key to building models that aid healthcare professionals in the diagnosis and treatment of anemia. Employing the Cross-Industry Standard Process for Data Mining (CRISP-DM), with its six phases, facilitates this endeavour. Our study compared Naïve Bayes, J48 Decision Tree, and Random Forest algorithms using RapidMiner's tools, evaluating accuracy, mean recall, and mean precision. The J48 Decision Tree outperformed the others, highlighting the importance of algorithm choice in anemia classification models. Furthermore, our analysis identified renal disease-related and chronic anemia as the most prevalent types, with a higher incidence among women. Recognizing gender disparities in the prevalence of anemia informs personalized healthcare decisions. Understanding demographic factors in specific types of anemia is crucial for effective care strategies
Predicting Smart Office Electricity Consumption in Response to Weather Conditions Using Deep Learning
No full textThis study investigates the intricate relationship between electricity consumption in smart office environments, temporal elements such as time, and external factors such as weather conditions. Using a data set that encompasses electrical consumption statistics, temporal data, and weather conditions, the research employs preprocessing, visualization, and feature engineering techniques. The predictive model for electric energy usage is constructed using deep learning architectures, including Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (Bi-LSTM), Gated Recurrent Unit (GRU), and Bidirectional Gated Recurrent Unit (Bi-GRU). Evaluation metrics reveal that the LSTM model outperforms others, achieving minimal Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). The study acknowledges the limitations of the data set, particularly when comparing electricity usage during work hours and outside working hours in a residential context. Future research aims to address these limitations, considering detailed meteorological data, missing data imputation, and real-time applications for broader applicability. The ultimate goal is to develop a predictive model that serves as a valuable tool for improving energy management in smart office settings, optimizing electricity usage, and contributing to long-term firm profitability