Jurnal RESTI (Rekayasa Sistem dan Teknologi Informasi)
Not a member yet
1071 research outputs found
Sort by
Deep Learning-Based Waste Classification with Transfer Learning Using EfficientNet-B0 Model
No full textRecycling of waste is a significant challenge in modern waste management. Conventional techniques that use inductive and capacitive proximity sensors exhibit limitations in accuracy and flexibility for the detection of various types of waste. Indonesia generates approximately 175,000 tons of waste per day, highlighting the urgent need for efficient waste management solutions. This study develops a waste classification system based on deep learning, leveraging the powerful EfficientNet-B0 model through transfer learning. EfficientNet-B0 is designed with a compound scaling method, which uniformly scales network depth, width, and resolution, providing an optimal balance between accuracy and computational efficiency. The model was trained on a dataset containing six classes of waste—glass, cardboard, paper, metal, plastic, and residue—totalling 7014 images. The model was trained using data augmentation and fine-tuning techniques. The training results show a test accuracy of 91.94%, a precision of 92.10%, and a recall of 91.94%, resulting in an F1-score of 91.96%. Visualization of predictions demonstrates that the model effectively classifies waste in new test data. Implementing this model in the industry can automate the waste sorting process more efficiently and accurately than methods based on inductive and capacitive proximity sensors. This study underscores the significant potential of deep learning models, particularly EfficientNet-B0, in industrial waste classification applications and opens opportunities for further integration with sensor and robotic systems for more advanced waste management solutions
Knowledge Management Foundation and Solutions Implementation in Indonesian Government Higher Educational Institution
No full textThe performance of XYZ, a Government Higher Educational Institution (GHEI) in Indonesia is assessed through two unintegrated applications. The 2023 target performance was missed due to miscalculations outside applications while transforming large data amounts. Thus, business intelligence (BI) serves as a knowledge management (KM) tool to integrate those applications to achieve XYZ's target. Because BI is costly and has a 70% failure rate of development plans, a research model was evaluated to look at the current XYZ innovation capability for successful BI adoption from the KM foundation and KM solution implementation. This study used a quantitative method, employing a questionnaire for 94 civil servants and the partial least squares-structural equation model (PLS-SEM) for data analysis. Results indicate in the KM foundation, organizational (O) negatively influences KM process application (KMP) (β = -0.292, Pv = 0.010) while KM infrastructure (I) and process (P) positively influence KMP, but KM technology (T) does not. In KM solutions, KMP is proven to be linked to innovation capability when KM systems are lacking. Hence, several activities are suggested to activate T through T, O, P, and I. The model validated 80% of the hypotheses, laying the groundwork for future studies into which aspects of T strengthen innovation capabilities in GHEI
Lightweight Models for Real-Time Steganalysis: A Comparison of MobileNet, ShuffleNet, and EfficientNet
No full textIn the digital age, the security of communication technologies is paramount, with cybercrime projected to reach $10.5 trillion annually by 2025. While encryption is vital, decrypted data remains vulnerable, prompting the exploration of steganography as an additional security layer. Steganography conceals data within digital media, but its misuse for cyberattacks—such as embedding malware—has highlighted the need for steganalysis, the detection of hidden data. Despite extensive research, few studies have explored lightweight deep learning models for real-time steganalysis in resource-constrained environments like mobile devices. This research evaluates MobileNet, ShuffleNet, and EfficientNet for such tasks, using the BOSSbase-1.01 dataset. Models were assessed based on accuracy, computational efficiency, and resource usage. MobileNet achieved the highest computational speed but with only 63.8% accuracy, falling short of practical application. ShuffleNet and EfficientNet performed at random-guessing levels with 50% accuracy, reflecting the challenges of steganalysis on mobile platforms. Future work aims to improve accuracy by integrating advanced preprocessing techniques, attention mechanisms, and hybrid architectures, as well as leveraging ensemble methods for improved detection. Data augmentation, transfer learning, and hyperparameter tuning will also be explored to optimize model performance. This study contributes by identifying these challenges and offering insights for future research, focusing on optimizing models and preprocessing techniques to enhance detection accuracy in resource-constrained environments
Comparative Evaluation of IndoBERT, IndoBERTweet, and mBERT for Multilabel Student Feedback Classification
No full textStudent feedback plays a crucial role in enhancing the quality of educational programs, yet analyzing this feedback, especially in informal contexts, remains challenging. In Indonesia, where student comments often include colloquial language and vary widely in content, effective multilabel classification is essential to accurately identify the aspects of courses being critiqued. Despite the development of several BERT-based models, the effectiveness of these models for classifying informal Indonesian text remains underexplored. Here we evaluate the performance of three BERT variants—IndoBERT, IndoBERTweet, and mBERT—on the task of multilabel classification of student feedback. Our experiments investigate the impact of different sequence lengths and truncation strategies on model performance. We find that IndoBERTweet, with a macro F1-score of 0.8462, outperforms IndoBERT (0.8243) and mBERT (0.8230) when using a sequence length of 64 tokens and truncation at the end. These findings suggest that IndoBERTweet is well-suited for handling the informal, abbreviated text common in Indonesian student feedback, providing a robust tool for educational institutions aiming for actionable insights from student comments
Rotation Double Random Forest Algorithm to Predict The Food Insecurity Status of Households
Get PDFThe ensemble tree method has been proven to handle classification problems well. The strength of the ensemble tree technique lies in the diversity and independence between each tree. Increasing the diversity of mutually independent decision trees improves the performance of the model. Various studies propose the development of ensemble tree-based models by forming algorithms that create decision trees that are formed independently of each other and have various inputs. These include random forest (RF), rotation forest (RoF), double random forest (DRF), and the latest is rotation double random forest (RoDRF). RoDRF rotates or transforms data with the intent of producing better diversity among the learning base. RoDRF applies the concept of variable rotation to trees based on the DRF algorithm. Random rotations or transformations on different feature subspaces produce different projections, leading to better generalization or prediction performance. This research aims to compare the performance of RoDRF with the RF, RoF, and DRF models on unbalanced data in cases of food insecurity. Class imbalance will be handled with two methods, namely EasyEnsemble and SMOTE-NC. The research results show that the DRF's model with EasyEnsemble techniques produces a model with the best performance among several algorithms tested. Although the resulting precision is 0.62274 and the AUC value is 0.68501, the model can predict each class equally. All algorithms with EasyEnsemble treatment have average AUC values significantly different from each other based on statistical test results. This research also used SHAP to explain variables that significantly contribute to the household's food insecurity status model
IoT-Based Irrigation System Using Machine Learning for Precision Shallot Farming
Get PDFDespite the massive production of shallots in Enrekang Regency, South Sulawesi Province, Indonesia, the cultivation method is still very conventional. Shallot cultivation is very challenging because it requires precision irrigation and pest prevention. In this research, we proposed a smart irrigation system to help farmers manage irrigation with more efficient water usage without hampering their pest prevention. The results of the system were three options: 1) no water needed, 2) water is required and is efficient for irrigation, and 3) water is required but it is not efficient for irrigation. We used Wireless Sensor Networks and IoT to collect yield parameters, designed a firebase database, and developed a mobile application and a web service embedded with a machine learning application. All applications interacted using the representational state transfer application programming interface. The proposed system architecture successfully gathered cropland data and distributed them to all applications within the system. Furthermore, we analyzed four supervised learning algorithms (decision trees, random forest, gradient boosting, and K-Nearest neighbor), and the random forest was deployed in the web service because it outperformed other algorithms with a precision of 94% and an AUC score of 0.90
An Analysis of Meteorological Data in Sumatra and Nearby using Agglomerative Clustering
Get PDFSumatra is one of the biggest and the second most crowded islands in Indonesia. Sumatra is also a place of abundance of tropical flora and fauna. This paper aims to cluster the cities in Sumatra and nearby based on the meteorology data. It implements Agglomerative hierarchical clustering and uses a daily time series dataset from 17 cities from 1 January 2010 to 31 December 2023. The dataset contains variables minimum temperature, maximum temperature, average temperature, humidity, sunshine duration, and average wind speed. The preprocessing data was dedicated to managing the missing values and data aggregation to create single-form data. The single-form data contains cities and meteorological variables used as an input for the clustering algorithm, i.e. K-Means, Fuzzy C-Means, K-Medoid, intelligent K-KMeans, and Agglomerative clustering. The Agglomerative clustering outperforms other methods (i.e. K-Means, Fuzzy C-Means, K-Medoid, and intelligent K-KMeans) and produces Silhouette scores of 0.11. The clusters are then analyzed to find their unique pattern. The cut-off when the number cluster is two, Agglomerative hierarchical clustering gathers Aceh, Sabang, Pekanbaru, Padang, and Padang Lawas in Cluster 1. Other cities, i.e., Nagan Raya, Batam, Jambi, Bandar Lampung, Medan, Pangkalpinang, Palembang, Bengkulu, Belitung, Tapanuli, Deli Serdang, and Nias are in Cluster 2. The results can be briefly explained that the characteristic of Cluster 1 has a higher average temperature, lower humidity, and lower sunshine duration than cities in Cluster 2. However, Cluster 1 has a lower average minimum temperature than Cluster 2. The pairs of cities which have the most similarities are (Aceh, Sabang), (Pekanbaru, Padang Lawas), (Nagan Raya, Nias), (Jambi, Palembang), (Bengkulu, Tapanuli), and (Medan, Deli Serdang). The annual trend in several cities shows that there exists an increasing trend in minimum temperature, rising sunshine duration, and decreasing wind speed. These are signs of climate change that need a proper handling
Hybrid Data Mining For Member Determination And Financing Prediction In Syariah Financing Saving And Loan Cooperatives
No full textSyariah Financing Saving And Loan Cooperatives (KSPPS) is an Islamic financial institution aimed at people who are on the lower middle scale to lift the economy of small communities through microfinancing programs. Problems that often occur in member recommendations to get KSPPS financing are often not on target. In addition, The amount of member financing is often problematic due to a lack of analysis, resulting in poor financing instalments. This research aims to present an analysis model for clustering and classification using hybrid data mining algorithms. This research method is using hybrid data mining Algorithms, namely K-Medoids, Naïve Bayes, and k-Nearest Neighbors (k-NN). This study uses the historical dataset of the last two years on KSPPS BMT Dadok Tunggul Hitam as a total of 70 data samples. The analysis parameters consist of income, business, residence Status, financing application, billing history, and balance amount. The best analysis Model will be obtained by comparing the results between Naïve Bayes with K-Medoids, and K-Nearest Neighbor (k-NN) with K-Medoids. The results of this research showed the best performance is using the hybrid Naïve Bayes data mining model with K-Medoids which has an accuracy of 90.91% for data split 70:30, while performance with K-fold cross-validation shows an accuracy of 93.49% using this algorithm. Overall, the results of this study can provide an effective analysis model to determine the status of the loan
Assessing User Experience and Usability in the OVO Application: Utilizing the User Experience Questionnaire and System Usability Scale for Evaluation
Get PDFThe OVO application, despite having a large user base in Indonesia, has received low ratings compared to other digital wallet apps on the Google Play Store and App Store. Users frequently complain about the user experience, which greatly affects their overall satisfaction. This study evaluates the user experience and usability of the OVO application using the User Experience Questionnaire (UEQ) and System Usability Scale (SUS). The UEQ results show that efficiency is excellent (1.55), while attractiveness, perspicuity, dependability, and stimulation are above average (1.56, 1.67, 1.33, and 1.16, respectively). However, the novelty aspect falls below average (0.64), indicating a need for improvement. The SUS score is 77.53, classifying the application as "Acceptable" with a "C" grade and an overall "Good" rating. Addressing identified shortcomings can enhance user experience and usability, ultimately improving user satisfaction. This study contributes valuable empirical data to the field, offering insights for researchers and practitioners in assessing the user experience and usability of mobile applications.The OVO application, despite having a large user base in Indonesia, has received low ratings compared to other digital wallet apps on the Google Play Store and App Store. Users frequently complain about the user experience, which greatly affects their overall satisfaction. This study evaluates the user experience and usability of the OVO application using the User Experience Questionnaire (UEQ) and System Usability Scale (SUS). The UEQ results show that efficiency is excellent (1.55), while attractiveness, perspicuity, dependability, and stimulation are above average (1.56, 1.67, 1.33, and 1.16, respectively). However, the novelty aspect falls below average (0.64), indicating a need for improvement. The SUS score is 77.53, classifying the app as "Acceptable" with a "C" grade and an overall "Good" rating. Addressing the identified shortcomings can enhance the user experience and usability, ultimately improving user satisfaction. This study contributes valuable empirical data to the field, offering insights for researchers and practitioners in assessing the user experience and usability of mobile applications
Ontology-based Conversational Recommender System for Recommending Camera
Get PDFThe camera is a product that has developed very quickly in terms of specifications and functions. In addition, the cameras available on the market are becoming increasingly varied, so customers need more time to find a camera that suits their needs. Currently, many recommender systems have been developed to assist users in finding suitable products, especially the conversational recommender system (CRS). CRS is a recommender system that recommends products through conversations between the user and the system. However, many developed CRS still forces users to have knowledge of the product's technical characteristics. In the real world, many people are not familiar with the technical features of products, especially cameras. People interact more easily with CRS by stating the camera function they want. In this study, we call that statement functional requirements. Therefore, we proposed a CRS for recommending cameras that interact with users using functional requirements. This CRS uses semantic reasoning techniques on ontologies. To evaluate system performance, we use two parameters, i.e., user satisfaction and recommendation accuracy. The evaluation results show that the accuracy of the recommendations is at a value of 82.35%, and the level of user satisfaction reaches 0.66. With these results, the system can provide recommendations accurately and satisfy users.
The camera is a product that has developed very quickly in terms of specifications and functions. In addition, the cameras available on the market are becoming increasingly varied, so customers need more time to find a camera that suits their needs. Currently, many recommender systems have been developed to assist users in finding suitable products, especially the conversational recommender system (CRS). CRS is a recommender system that recommends products through conversations between the user and the system. However, many developed CRS still forces users to have knowledge of the product's technical characteristics. In the real world, many people are not familiar with the technical features of products, especially cameras. People interact more easily with CRS by stating the camera function they want. In this study, we call that statement functional requirements. Therefore, we proposed a CRS for recommending cameras that interact with users using functional requirements. This CRS uses semantic reasoning techniques on ontologies. To evaluate system performance, we use two parameters, i.e., user satisfaction and recommendation accuracy. The evaluation results show that the accuracy of the recommendations is at a value of 82.35%, and the level of user satisfaction reaches 0.66. With these results, the system can provide recommendations accurately and satisfy users