Computer Engineering and Applications Journal (ComEngApp, Universitas Sriwijaya)
Not a member yet
102 research outputs found
Sort by
Point of Interest (POI) Recommendation System using Implicit Feedback Based on K-Means+ Clustering and User-Based Collaborative Filtering
No full textRecommendation system always involves huge volumes of data, therefore it causes the scalability issues that do not only increase the processing time but also reduce the accuracy. In addition, the type of data used also greatly affects the result of the recommendations. In the recommendation system, there are two common types of data namely implicit (binary) rating and explicit (scalar) rating. Binary rating produces lower accuracy when it is not handled with the properly. Thus, optimized K-Means+ clustering and user-based collaborative filtering are proposed in this research. The K-Means clustering is optimized by selecting the K value using the Davies-Bouldin Index (DBI) method. The experimental result shows that the optimization of the K values produces better clustering than Elbow Method. The KMeans+ and User-Based Collaborative Filtering (UBCF) produce precision of 8.6% and f-measure of 7.2%, respectively. The proposed method was compared to DBSCAN algorithm with UBCF, and had better accuracy of 1% increase in precision value. This result proves that K-Means+ with UBCF can handle implicit feedback datasets and improve precision
Query Reformulation for Indonesian Question Answering System Using Word Embedding of Word2Vec
No full textQuery reformulation is one of the tasks in Information Retrieval (IR), which automatically creates new queries based on previous queries. The main challenge of query reformulation is to create a new query whose meaning or context is similar to the old query. Query reformulation can improve the search for relevant documents for Open-domain Question Answering (OpenQA). The more queries are given to the search system, and the more documents will be generated. We propose a Word Predicted and Substituted (WPS) method for query reformulation using a word embedding word2vec. We tested this method on the Indonesian Question Answering System (IQAS). The test results obtained an E-1 value of 81% and an E-2 value of 274%. These results prove that the query reformulation method with WPS and word-embedding can improve the search for potential IQAS answers
Parking System Optimization Based on IoT using Face and Vehicle Plat Recognition via Amazon Web Service and ESP-32 CAM (Case Study: Institut Teknologi Sumatera)
No full textToday\u27s technology has developed rapidly. One application of technology is in the parking lot. Most parking lots in Indonesia can already recognize the vehicle plate image, but it is hoped that it can be even better by applying Internet of Things (IoT) technology that is integrated with facial recognition images. One of the parking problems is in the parking lot at the Sumatran Institute of Technology, where checking is still done manually by security officers. This of course will take time and the level of security is also not good, because when you enter there is no checking. Checks are only carried out at the time of exit and the officer who checks is not necessarily the same and memorized as the owner of the vehicle. The addition of this facial image recognition feature is expected to increase the security of the parking system. Facial image recognition can be assisted by Cloud services from Amazon Image Recognition. With this service, no training data is required. The system developed is only a prototype. The developed parking system can recognize facial images and vehicle license plates with 2 cameras using the ESP32-Cam when entering and exiting the parking lot. The use of the ESP32-cam can recognize facial images both during the day and at night. The results obtained by the system can work effectively with an increase of 21%
Comparative Analysis Multi-Robot Formation Control Modeling Using Fuzzy Logic Type 2 – Particle Swarm Optimization
No full textMulti-robot is a robotic system consisting of several robots that are interconnected and can communicate and collaborate with each other to complete a goal. With physical similarities, they have two controlled wheels and one free wheel that moves at the same speed. In this Problem, there is a main problem remaining in controlling the movement of the multi robot formation in searching the target. It occurs because the robots have to create dynamic geometric shapes towards the target. In its movement, it requires a control system in order to move the position as desired. For multi-robot movement formations, they have their own predetermined trajectories which are relatively constant in varying speeds and accelerations even in sudden stops. Based on these weaknesses, the robots must be able to avoid obstacles and reach the target. This research used Fuzzy Logic type 2 – Particle Swarm Optimization algorithm which was compared with Fuzzy Logic type 2 – Modified Particle Swarm Optimization and Fuzzy Logic type 2 – Dynamic Particle Swarm Optimization. Based on the experiments that had been carried out in each environment, it was found that Fuzzy Logic type 2 - Modified Particle Swarm Optimization had better iteration, time and resource and also smoother robot movement than Fuzzy Logic type 2 – Particle Swarm Optimization and Fuzzy Logic Type 2 - Dynamic Particle Swarm Optimization
Effect of Genetic Algorithm on Prediction of Heart Disease Stadium using Fuzzy Hierarchical Model
No full textThe Fuzzy Hierarchical Model method can be used to predict the stage of heart disease. The use of the Fuzzy Hierarchical Model on complex problems is still not optimal because it is difficult to find a fuzzy set that provides a more optimal solution. This method can be improved by changing the membership function constraints using Genetic Algorithm to get better predictions. Tests carried out using 282 heart disease patient data resulted in a Root Mean Squared Error (RMSE) value of 0.55 using the best Genetic Algorithm parameters, including population size of 140, number of generations of 125, and a combination of cross-over rate and mutation rate of 0.4 and 0.6 whereas the RMSE value generated by the Fuzzy Hierarchical Model before being optimized by the Genetic Algorithm was 0.89. These results indicate an increase in the predictive value of the Fuzzy Hierarchical Model after being optimized using the Genetic Algorithm
Identification of Stunting Disease using Anthropometry Data and Long Short-Term Memory (LSTM) Model
No full textChildren with unbalanced nutrition are currently crucial health issues and under the spotlight around the world. One of the terms for malnourished children is stunting. Stunting is a disease of malnutrition found in children aged under 5 years; as many as 70% of stunting sufferers are children aged 0-23 months. There are several ways to diagnose stunting, one of which is using stunting anthropometry. Stunting anthropometry can measure the physique of children so that some of the features that characterize the presence of stunting can be identified. Features resulted from the stunting anthropometry cover age, height, weight, gender, upper arm circumference, head size, chest circumference, and hip fat measurement. The process of identifying stunting can be simplified using an intelligent system called the Computer-Aided Diagnosis (CAD) system. CAD system contains 2 main processes, namely preprocessing and classification. Preprocessing includes normalization and augmentation of data using the SMOTE method. The classification process in this study uses the LSTM method. LSTM is a modification of the Recurrent Neural Network (RNN) method by adding a memory cell so that it can store memory data for a long time and in large quantities. The results of this study compare between the results of models that apply preprocessing and the one without preprocessing. The model that only uses LSTM has the best accuracy of 78.35%; the model with normalization produces an accuracy of 81.53%; the model that uses SMOTE produces an accuracy of 81.66%; and the model that uses normalization and SMOTE produces the best accuracy of 85.79%
Analysis and Implementation of Augmented Reality Using Markerless and A-Star Algorithm (Case Study: Gedung Kuliah Umum ITERA)
No full textInstitut Teknologi Sumatera is a public university in the province of Lampung. Institut Teknologi Sumatera (ITERA) has many buildings, including Gedung Kuliah Umum (GKU). GKU is the largest and widest lecture building in ITERA. GKU has four floors, where each floor has many rooms in it with different functions in each room. As the largest building in ITERA, GKU is often used for various events, including CPNS exams, new student admissions, or for visits from other campuses. Due to the size of this building, this allows visitors from outside ITERA to GKU to experience problems in terms of time to ask questions and difficulty finding various spaces in the GKU Building. This research uses Augmented Reality technology to help make it easier for visitors from outside ITERA to find space quickly and precisely. In its development using several tools, including the ARWaKit SDK. This framework is used on devices with the IoS operating system. In its implementation, it requires a camera on a smartphone to capture existing images and convert them into cyberspace. In the ARWayKit framework, Azure Spatial Anchors have been used which can be used to carry out the mapping process as a markerless method and to optimize the distance from the user\u27s position to the destination location, the a-star algorithm is used. The results obtained from the Variation-2 test were 91.6%
Segmentation of the Lungs on X-Ray Thorax Image with CNN Architecture U-Net
No full textLungs are one of the most important parts of the human body. They are very susceptible to various disorders and diseases. For this reason, it is necessary to detect or diagnose the lungs. In this study, we present a method for lung segmentation using the CNN method U-Net architecture. The initial stage was preprocessed did a 1-1 correspondence to equalize the amount of training data and testing data and resized the image so all images have the same size. The process continued with the CLAHE (Contrast Limited Adaptive Histogram Equalization), and after that, the segmentation process was carried out according to the method. This study used a dataset from the Kaggle website. The results used the CNN method of the U-Net architecture in data get an average accuracy of 91.68%, sensitivity 92.80%, and specificity 89.15%, precision 95.07, and F1-Score 93. 92%. Based on the performance evaluation results, it was concluded that the method proposed in the study is great and valid in the lungs segmentation on X-Ray Thorax images
Inter Patient Atrial Fibrillation Classification Using One Dimensional Convolution Neural Network
No full textAtrial fibrillation is the most common type of arrhythmia. The process of detecting AF disease is quite difficult. This is because it is necessary to detect the presence or absence of a P signal wave in the ECG signal. However, this method requires special expertise from a cardiologist. Several literatures have proposed an automatic ECG classification system. However, the intra-patient paradigm does not simulate real-world scenarios. One of the challenges in the inter-patient paradigm is the morphological differences between one subject and another. In order to overcome the problems that arise in the automatic classification of ECG signal patterns a deep learning approach was proposed. This study proposes the classification process of atrial fibrillation in the inter-patient paradigm using a one-dimensional convolutional neural network architecture. The test is divided into two cases: two labels (Normal and AF) and three labels (Normal, AF and Non-AF). In the case of two test labels with an inter-patient scheme, the performance was 100% for all test metrics (accuracy, sensitivity, precision, and F1-Score). However, in the three-label case, the model\u27s performance decreased to 85.95, 70.02, 72.50, 71.19 for accuracy, sensitivity, precision and F1-Score, respectively
Brahmi Script Classification using VGG16 Architecture Convolutional Neural Network
No full textMany Indonesians have difficulty reading and learning the Brahmi script. Solving these problems can be done by developing software. Previous research has classified the Brahmi script but has not had an output that matches the letter. Therefore, letter classification is carried out as part of the process of recognizing Brahmi script. This study uses the Convolutional Neural Network (CNN) method with the VGG16 architecture for classifying Brahmi script writing. Training results from various amounts of image data. Smooth model. The requested image data is a 224x224 binary image. This study has the highest quality, accuracy is 96%, highest recall is 98% and highest precision is 98%