Kinetik: Game Technology, Information System, Computer Network, Computing, Electronics, and Control
Not a member yet
424 research outputs found
Sort by
Forecasting Model for Lighting Electricity Load with a Limited Dataset using XGBoost
Get PDFEnergy forecasting is an important application of machine learning in renewable energy because it is used for operational, management, and planning purposes. However, using the electricity load dataset during COVID-19 is a research challenge in the forecasting model due to the limited dataset and non-stationarity. This paper proposes an extreme gradient boosting (XGBoost) forecasting model for a limited dataset. Forecasting models require large amounts of data to create high-accuracy models. We conduct research using the PT Biofarma office electricity usage dataset for eight months during the COVID-19 period. Because office activities were limited during the pandemic, the datasets obtained were few. Several methods are used for modeling limited datasets, namely XGBoost, multi-layer perceptron (MLP), autoregressive integrated moving average (ARIMA), and long short-term memory (LSTM). We have conducted several experiments using a limited dataset with these four methods. The test results with the t-test show that the electricity load data for work-from-office (WFO) and work-from-home (WFH) periods have a significant average difference. Then the test results with the augmented DickeyโFuller (ADF) test show that our data is non-stationary. Compared to the benchmark method, the XGBoost method shows the best forecasting performance with mean absolute percentage error (MAPE), root mean squared error (RMSE), mean absolute error (MAE), and R2 of 0.48, 5.00, 3.09, and 0.61 respectively
A Systematic Review of Artificial Intelligence in Assistive Technology for People with Visual Impairment
Get PDFRecent advances in artificial intelligence (AI) have led to the development of numerous successful applications that utilize data to significantly enhance the quality of life for people with visual impairment. AI technology has the potential to further improve the lives of visually impaired individuals. However, accurately measuring the development of visual aids continues to be challenging. As an AI model is trained on larger and more diverse datasets, its performance becomes increasingly robust and applicable to a variety of scenarios. In the field of visual impairment, deep learning techniques have emerged as a solution to previous challenges associated with AI models. In this article, we provide a comprehensive and up-to-date review of recent research on the development of AI-powered visual aides tailored to the requirements of individuals with visual impairment. We adopt the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) methodology, meticulously gathering and appraising pertinent literature culled from diverse databases. A rigorous selection process was undertaken, appraising articles against precise inclusion and exclusion criteria. Our meticulous search yielded a trove of 322 articles, and after diligent scrutiny, 12 studies were deemed suitable for inclusion in the ultimate analysis. The study's primary objective is to investigate the application of AI techniques to the creation of intelligent devices that aid visually impaired individuals in their daily lives. We identified a number of potential obstacles that researchers and developers in the field of visual impairment applications might encounter. In addition, opportunities for future research and advancements in AI-driven visual aides are discussed. This review seeks to provide valuable insights into the advancements, possibilities, and challenges in the development and implementation of AI technology for people with visual impairment. By examining the current state of the field and designating areas for future research, we expect to contribute to the ongoing progress of improving the lives of visually impaired individuals through the use of AI-powered visual aids
PID Controller for DC-DC Converter under Dynamic Load Change in Photovoltaics based Low-Voltage DC Microgrid
Get PDFToday, DC Microgrid gain more attraction due to increasing electronic digital devices application such smart-phones, smart-tvs, and other digital apparatus which are operated in DC form. In the common grid, the electric power from AC source is converted to DC voltage for powering the digital devices as load. Hence, there are power conversions from AC-DC and potentially loss energy during conversions. DC Microgrid consisted of sources, loads, distribution lines and energy storages. In small capacity DC Microgrid, the stability of the system is vulnerable by dynamic load change. During load demands fluctuations, the DC Microgrid voltage is also dynamically fluctuated and can reach over the designated rate. To solve this problem, the PID controller is introduced in the DC-DC converter for maintaining the voltage rate at designated value regardless the load demands. In this paper, the DC Microgrid is consisted of photovoltaics as DC sources, XL-6019 as DC-DC converter, Arduino as controller, voltage and current sensors, distribution lines and loads. The proposed method is evaluated via experimental results. The responses of the proposed method in the DC Microgrid system are presented, evaluated, discussed, and compared between with and without applied method. The experimental results indicate that the proposed method has ability to reduce the voltage profile fluctuations during load demands changes and in short time
Threat Construction for Dynamic Enemy Status in a Platformer Game using Classical Genetic Algorithm
Get PDFDigital game genre such as Action-Platformer is widely popular among buyers on a platform like Steam. The non-playable character enemies in the game are important in action games. Unfortunately, they usually have static attributes like health points, damage, and enemy movement. Using the combination of procedural content generation and dynamic difficulty adjustment with a classical genetic algorithm, we drive the threat value of a platform to construct the enemy status, resulting in more dynamic enemies. We use the threat value as an input parameter calculated from the enemiesโ stats in every platform, such as total damage that the enemy might produce, the playerโs health point, and the enemyโs movement speed. We conclude that using a classical genetic algorithm may produce dynamic enemy status through the desired threat or danger set by the game designer as an input parameter. Moreover, the game designer may limit the generation with constraints
Enhancing Accuracy on Chronic-Kidney Disease Detection Using Machine Learning with Technique of Resampling and Missing Value Treatment
Get PDFChronic kidney disease is one of the deadliest diseases in the world. It is important to identify chronic kidney disease at an early stage, so that treatment and prevention can be carried out early. This study used linear interpolation method to treat the missing values, resampling using SMOTE method, and several feature selection methods, such as Pearsonโs correlation coefficient and Principal component analysis. For the classification methods, Support Vector Machine and Logistic Regression were used to build prediction models for chronic kidney disease based on dataset on UCI Machine Learning. To measure the performance of the model, several test scenarios were tested out so it can be compared to the previous research on the detection of chronic kidney disease, which is used as a benchmark for this study. The best result from the experiment is obtained from the scenario of resampling using SMOTE and feature selection using Principal Component Analysis with averaged accuracy, precision, and f1-score respectively are 98,8%, 100%, dan 98,77%
Combination of Term Weighting with Class Distribution and Centroid-based Approach for Document Classification
Get PDFA text retrieval system requires a method that is able to return a number of documents with high relevance upon user requests. One of the important stages in the text representation process is the weighting process. The use of Term Frequency (TF) considers the number of word occurrences in each document, while Inverse Document Frequency (IDF) considers the wide distribution of words throughout the document collection. However, the TF-IDF weighting cannot represent the distribution of words to documents with many classes or categories. The more unequal the distribution of words in each category, the more important the word features should be. This study developed a new term weighting method where weighting is carried out based on the frequency of occurrence of terms in each class which is integrated with the distribution of centroid-based terms which can minimize intra-cluster similarity and maximize inter-cluster variance. The ICF.TDCB term weighting method has been able to provide the best results in its application to SVM modeling with a dataset of 931 online news documents. The results show that SVM modeling had accuracy of 0.723, outperforming the use of other term weightings such as TF.IDF, ICF & TDCB
Comparison of CNNโs Architecture GoogleNet, AlexNet, VGG-16, Lenet -5, Resnet-50 in Arabic Handwriting Pattern Recognition
Get PDFThe Arabic script is written from right to left and consists of 28 characters, with no capital or lowercase letters. The Arabic script has several orthographic and morphological properties that make handwriting recognition of the Arabic script challenging. In addition, one of the biggest challenges in recognizing Arabic script patterns is the different handwriting styles and characters of each person's writing. The authors propose a study to compare the accuracy of handwriting pattern recognition in Arabic script which has been done previously by comparing five CNN architectures, namely GoogleNet, AlexNet, VGG-16, LeNet-5, and ResNet-50. Considering that previous research has not obtained excellent accuracy. The number of datasets used is 8400 image data and the most optimal comparison of testing and training data is 80:20. Based on the research that has been done, there are several things that the author can conclude. The model is made using 64 filters for each convolution layer because the optimal size is used for 5 architectures, kernel size is 3x3, neurons is 128, dropout weight is 50% to reduce overfitting, learning rate is 0.001, image size is 64x64, the normalization method with the ReLU activation function, and 1-dimensional input image (grayscale), and with a comparison of testing and training data of 80:20. The VGG-16 architectural model is the architecture that gets the highest score, namely 83.99%. This can have good potential to be developed as a medium for learning Arabic script
Public Opinion Analysis of Presidential Candidate Using Naรฏve Bayes Method
Get PDFElections for president and vice president will take place in 2024. Heading into the election, promoted candidates were vying for public sympathy. People often discussed as presidential candidates are Anies Baswedan, Ganjar Pranowo, and Prabowo Subianto. Therefore, we need a way to predict potential candidates and voter demographics from public opinion on Twitter using sentiment analysis. One of his methods commonly used to classify sentiment analysis is Naive Bayes. This study used the naive Bayes classifier and the TF-IDF extraction function to add weights to the text. Use the scikit-learn Python library to help determine the polarity of negative and positive sentiment classes in your dataset. The datasets used were Twitter datasets acquired from October to December 2022, for a total of 15,000 datasets. The best test scenario obtained by splitting the test and training data is 70% test data and 30% training data, with the highest accuracy generated from the 95% Ganjar dataset. Using the Anies, Ganjar, and Prabowo test data, the positive mood scores for each candidate were 833, 77, and 524, respectively, while the negative mood scores were 637, 1423, and 976, respectively. The test was performed using a confusion matrix and k-fold cross-validation, and the best results were obtained on the Ganjar data set. That is a confusion matrix of 94.93% and a k-fold cross-validation of 94.46%. The lowest f1-score for the positive class is 67% for the Anies dataset and 27% for the negative class for the Ganjar dataset
Risk Management using COBIT 5 for Risk : A Case Study on Local Government in Indonesia
Get PDFBP4D (Regional Development Planning, Research and Development Agency) Bondowoso utilizes information technology to support its duties and functions, one of which is SIPD (Sistem Informasi Pemerintah Daerah). SIPD provides many benefits and conveniences such as improving the quality of public services, transparency, improving bureaucratic accountability, but in its implementation SIPD can also pose dangerous risks both from processes involving the system and the system itself. These risks can disrupt BP4D Bondowoso's business processes and cause various losses. To protect BP4D Bondowoso from losses caused by risk, risk management is carried out using the relevant framework, namely COBIT 5 Enabling Process and COBIT 5 for Risk with the APO12 risk management process. Data were collected by interview and distributing questionnaires. Fifty-one risks were identified in the implementation of SIPD at BP4D Bondowoso consisting of 48 negative risks and 3 positive risks. The risks found dominate the type of IT Benefit / Value Enablement and the category of regulatory compliance. Identified 3 very high risks in the category of regulatory compliance and software. Overall risk dominates the medium rating, which is 28 risks and the high risk consists of 20 risks. The negative risk response is dominated by mitigate, which is 33 risks
Sensor Fusion using Model Predictive Control for Differential Dual Wheeled Robot
Get PDFEvery mobile robot mission starts with the robot being moved to the task site. From there, the robot executes its tasks. A control system is required to move the mobile robot's actuator (which may be in the shape of wheels or legs) and comprehend the environment around the robot to perform these movements (perception). This research aims to develop a technique to control a robotโs movement while detecting obstacles and distances toward an object. The robot is equipped with LIDAR and a camera to perform these tasks. The control is divided into two major parts, low-level and high-level controller. As part of a low-level controller robot, the Model Predictive Control (MPC) method is proposed to help with the control of the wheel while the Artificial Neural Network (ANN) approach to use in this study to identify obstacles and the Convolutional Neural Network (CNN) method for detecting objects, both ANN and CNN as a control for high-level part of the robot. The results of this study can prove that CNN can help detect existing objects with a value of 45% for detecting some objects. The obtained result from the MPC method, which has been combined with an ANN as an obstacle detector, is that the smaller the horizon value, the shorter the time needed to reach the desired coordinates with the result being 45 seconds