Various approaches have been proposed to detect code smells, including machine learning models. However, there are still challenges to improving detection accuracy and selecting appropriate quality metrics. This research introduces the use of Relevance Vector Machine (RVM) for code smell detection, especially Magic Value smells, which often affect source code readability and maintainability in a software development industry. This approach utilizes source code analysis using two source code representations (raw source code and token stream) and feature extraction using two text mining techniques (Bag of Words/BoW and Term Frequency-Inverse Document Frequency/TF-IDF). The experiment involves analyzing 983 records of labeled source code datasets. The results show that RVM can detect code smells with a very high level of accuracy, reaching 99.9% on some kernel configurations, also nearly perfect precision, and F1 scores. This paper contributes to introducing the use of RVM for detecting code smells in source code, explores source code representation and feature extraction techniques, and shows that RVM can achieve high detection accuracy.

Mining Geology is one of the study programs provided by Indonesian Vocational Schools that explores surface mining. Mining is a work in hazardous environments that poses numerous risks. The use of heavy equipment is a frequent cause of mining accidents. Therefore, enhancing vocational students' understanding of heavy equipment safety is essential. However, having heavy equipment for learning its safety is costly and complex maintenance, demanding substantial financial investment from schools. Moreover, many vocational schools still use traditional learning media that cannot provide an immersive experience in learning. Thus, this study developed an immersive and engaging virtual reality (VR)-based learning media for vocational students learning heavy equipment safety. The validity and feasibility of the learning media contents were evaluated by experts. The result concluded that the contents presented are valid and have good feasibility. Moreover, the learning media was tested in the learning process and the usability was assessed by vocational students. The result concluded that the learning media demonstrates significant potential to improve the educational experience in vocational schools, leading to better learning outcomes and increased student engagement.

This paper presents the implementation of forecasting of photovoltaic (PV) power and stored energy on small-scale pumped hydro storage (PHS) systems. The proposed forecasting approach considers the results of predicting solar irradiance, ambient temperature, and rainfall. Prediction of these three parameters was done using a one-month weather dataset that was split into 80% for training and 20% data for testing and validation. The prediction algorithm used is the bidirectional long short-term memory (LSTM) method. Furthermore, PV power and stored energy were calculated using power and energy models of the photovoltaic-pumped hydro storage system modified by considering the prediction results of solar irradiance, ambient temperature, and rainfall. The proposed forecasting approach was simulation-tested on a small-scale photovoltaic-pumped hydro storage system with a capacity of PV is 2 KW, as well as a capacity of the upper reservoir is 5 m3, The performance of the forecasting model is done by measuring the mean square error (MSE) and the mean absolute error (MAE) values. From the simulation test, the results obtained that the suggested approach produces PV power forecasting performance with an MSE of 8.942 watts and MAE of 0.044 watts, and excess power forecasting performance with an MSE of 81.203 watts and MAE of 0.075 watts. The stored energy forecasting performance for MSE and MAE parameters are$2.2\times 10^{-7}$Wh and$7.35\times 10^{-6}$Wh, respectively.

This paper presents neuromuscular disease classification using seventeen time-domain feature extraction, feature reduction, and a machine learning method. Electromyography (EMG) signals were collected from one of the neuromuscular diseases i.e. Parkinson Disease (PD). One of typical PD stages (classes) namely healthy, possible, probable, and definite were used in this study. Ten Subjects of each class participated in this study. The 10 subjects of the healthy class were collected from the University student and the other 10 subjects for each possible, probably, and definite class were collected from Kariadi General Hospital in Semarang, Indonesia. The EMG signals of each subject were proceeded with 17 time-domain feature extraction and followed by the dimensional feature reduction based on Principal Component Analysis (PCA). The total 17 features of each class were reduced up to 4 new features using PCA namely Principal Component (PC). A pair combination of PCs was used for training and testing using Support Vector Machine (SVM). A classification results show that the new features of PCA increased the classification accuracy.

Personalized learning paths are designed to optimize learning time and improve student learning performance by providing an appropriate learning sequence based on the unique characteristics of each student. A common method for constructing personalized learning paths is based on the student's knowledge but disregards the student's interest in the subject matter. This research employs a deep learning and rule-based approach to recommend suitable material based on the topic's difficulty, student interest, and knowledge level. The difficulty level of the topic is predicted using deep learning. A questionnaire is used to determine the level of student interest, which is then processed using a rule-based approach to generate a learning path. Modeling a dynamic learning path requires measuring student knowledge in each topic and updating the learning path accordingly. Comparing the learning outcomes of students who utilized conventional e-learning versus those who followed a personalized learning path constitutes the evaluation. The results demonstrated that students scored 29% higher, or 15.06 points, than those who utilized conventional e-learning.

Point cloud object classification has gained attention from many researchers since the emergence of public dataset like ModelNet and ShapeNet, which contains full surface objects. However, in practice, objects captured using LiDAR are only partially covered in the scanned area, making such a task burdensome. Here, we proposed a solution to overcome those problems. It is a novel fuzzy convolutional inference (FuzzConv) incorporated with depthwise over-parameterization (DOConv). Instead of applying raw data, the point clouds are transformed into a 3D voxel. We utilized EfficientNet as our backbone and modified the Mobile inverted Bottleneck Convolution (MBConv) with DOConv. In the last fully connected (FC) layer, we added the FuzzConv layer as an inference before feeding the feature map to the output layer. Consequently, to validate the performance of our model, we undertake an evaluation with multiple classifications in ModelNet10, ModelNet40, and our core dataset, the point cloud of human poses. Accuracy, loss, number of parameters, loss, precision, and F1-scores are employed as performance indicators. As a result, our model achieved top performance regarding the accuracy and loss value for the primary dataset, 83 % and 0.56, for ModelNet10 88.1 % and 0.56, and ModelNet40 74.1 % and 1.15.

Understanding pedestrian behaviour when crossing the road is an important key to the development of autonomous vehicles. Because pedestrians are considered Vul-nerable Road Users (VRUs), they are likely to be killed if they are involved in an accident. To ensure their safety, it is then necessary to predict the pedestrian's intention based on their behaviour. In this experiment, we propose head pose observation for predicting their intention, by observing pedestrians' head pose data, we can then predict their intention to cross the road. To achieve this purpose, we use human head detector and head pose extraction feature, and the resulting yaw, pitch and roll as three head pose features. To select the most optimal feature is important for predicting pedestrian intention, then we make 7 combination scenarios based on these three features and compare it with the same model. Based on this scenario, it is proved that all these three data are optimal to observe pedestrian intention. There are three behavioural annotation that have been used, there are crossing, not crossing and will crossing. We derive will crossing from the annotation looking and not crossing while waiting at the roadside. Prediction of pedestrian behaviour is done by using LSTM model, and the resulting precision on crossing and not crossing with 0.98, while will crossing is 0.94.

In digital development 4.0, store brands are very important. The problem in this research is the lack of consumer trust to buy quality goods in e-commerce store accounts so that it affects consumer satisfaction. This study aims to address this question feedback from the problems of the customer, then on the other hand a questionnaire with the PLS-SEM (Partial Least Squares Structural Equation Modeling) model to determine the dimensions of the variables selected according to customer experience. To achieve this aim, both negative and positive customer comments were compiled to assess customer satisfaction, employing a comparative analysis method through Naive Bayes algorithm. The overarching goal was to achieve optimal results and extract valuable insights regarding the determinants that influenced customer satisfaction within the domain of online transactions. This research also has an impact on buyers so they can have an understanding of the factors that support trust in customer satisfaction, so that individuals do not hesitate in making purchasing decisions to shop online. The results showed that the algorithm initially recorded a modest accuracy score of 0.37. Meanwhile, after implementing hyperparameter tuning, the accuracy increased significantly to 0.62. In the aspect of Smart PLS questionnaire analysis, a standardized Normed Fit Index (NFI) of 0.707 was recorded, which was slightly below the established threshold of 0.90. The standardized root mean square residual (SRMR) was measured at 0.071, falling below the specified value of 0.08, indicating a commendable model fit. However, the RMS theta value at 0.240 exceeded the threshold of 0.102.

Virus SARS-Cov-2 causing Covid-19 spreads quickly and brings high risks to transmissions. The government to rule strictly to arrange strategies to minimize interactions through School-From-Home (SFH) policy. Unfortunately, the school closure is the potential to hamper deliveries of education services and may entail destructive impacts to quality education performance. There must be a consideration to school reopen safely during the pandemic.The objective of the research is to produce a model of Covid-19 spreads to analyze the readiness of school to reopen. This study adopts a SEIR model to predict the spread of Covid-19 using dataset from 23 March through 31 December 2020. The best model is selected from the one having the least error and adopted to predict the spread in the next 100 days starting from 01 January 2021 through 10 April 2021.Clustering was then implemented to acquire the character’s proximity in each area using K-Means algorithm. While unsupervised fuzzy was picked out to seize the phenomenon of the dynamic as Covid-19 spread as a basis to decision making on school reopen safely during the pandemic. These whole concepts will serve the decision making effectively and intelligently by generating a better estimation.This study resulted in a Covid-19 spread model with an average error of 0.2% based on the RMSLE calculation.

The use of cameras to monitor the elderly daily living activities might cause inconvenience related to the privacy issues. Thus, another sensor namely LiDAR which generates point cloud data is used to support the monitoring process. This study is aimed at segmenting human and ground data from LiDAR point cloud to obtain human data. A segmentation process using a normal vector search approach for each point perpendicular to its plane using Principal Component Analysis (PCA) assisted by k-dimensional Tree Nearest Neighbor (kdTree-NN) & Singular Value Decomposition (SVD) is proposed and successfully implemented. KITTI dataset containing of 54 frames in which a human is walking towards the LiDAR sensor as a scene scenario was used. The trend of the number of raw data points increased by 12.58%. Furthermore, the trend in the number of data points segmented representing human also increased by 233.13%. Meanwhile, the data points segmented representing ground decreased by 6.36%. This is because the closer human walking to LiDAR, the wider the blank spot behind the human object is. Consequently, data points representing human increased significantly, reducing the number of data points representing the ground. The point clouds which both represent the human and the ground were successfully segmented. Therefore, the point cloud representing human was successfully obtained to be used in further research.

The need test data on facial expression recognition for Indonesian people are not easily obtained and not available in existing databases. Race and ethnic differences can affect the results of facial expression recognition. Testing on facial expressions using untrained data will give more objective results. The purpose of this study is to utilize the Acquiface interface feature in the game Expressions Hunter to get data in the form of images of facial expressions from Indonesian players. Finite State Machine is used to provide behavior to the enemy according to a user action to collect facial expressions by hunting and killing enemies. Acquiface interface is a prototype of one of the features in the game Expression Hunter which functions to record user expressions in accordance with the collection of expressions obtained by users after killing regular enemy monsters or enemy Bosses. RADIATE facial stimulus sets are used as stimuli to generate facial expressions according to the type of expression. The result of recording is a facial expression video file which is then extracted into a sequence of facial expressions. The final image of facial expression is selected from the most expressive sequence of facial expressions (having peak expression).

Emotion recognition based on physiological signal has become an important issue among researchers nowadays. It is because many studies have proven that emotion condition, especially in elderly, has influenced the physical condition significantly. Nevertheless, there are still few studies which discuss and explores emotion recognition based on SpO2and Pulse Rate Signals. This paper proposed emotion recognition of three basic emotions of elders, such as happy, sad and angry based on those physiological signals. Window size segmentation that was used to extract both physiological signals was 15 second. Then, statistical feature extraction method was used to obtain the features of SpO2and Pulse Rate (PR). Support Vector Machine (SVM) with selecting the best of C and γ parameters and the most optimal K parameters of k-Nearest Neighbors (k-NN) method were used to classify the extracted features which were tested in several scenarios: classification using SpO2, using PR and using SpO2-PR features. The result showed that SVM achieved the best accuracy (72.86%) and precision (71.30%) compared to k-NN. Furthermore, combining the features of both physiological signals could improve the accuracy and precision scores more than 3.70% compared to the single physiological signal. This result provides information of emotion recognition in term of SpO2and PR signals which can be better detected by combining the features of both physiological signals. Moreover, the optimal C and γ parameters of SVM and K-value of k-NN can be implemented to achieve better classification result.

In this paper, we develop a pattern recognition system to detect weather an infant is crying or not just by using his facial feature. The system must first detect the baby face by using the Haar-like feature, then find the facial component using trained active shape model (ASM). The extracted feature then fed to Chaotic Neural Network Classifier. We designed the system so that when the testing pattern is not a crying baby the system will be chaotic, but when the testing pattern is a crying baby face the system must switch to being periodic. Predicting whether a baby is crying based only on facial feature is still a challenging problem for existing computer vision system. Although crying baby can be detected easier using sound, most CCTV don't have microphone to record the sound. This is the reason why we only use facial feature. Chaotic Neural Network (CNN) has been introduced for pattern recognition since 1989. But only recently that CNN receive a great attention from computer vision people. The CNN that we use in this paper is the Ideal Modified Adachi Neural Network (Ideal-M-AdNN). Experiments show that Ideal-M-AdNN with ASM feature able to detect crying baby face with accuracy up to 93%. But nevertheless this experiment is still novel and only limited to still image.

Fixed order neural networks (FONN), such as high order neural network (HONN), in which its architecture is developed from zero order of activation function and joint weight, regulates only the number of weight and their value. As a result, this network only produces a fixed order model or control level. These obstacles, which affect preceeding architectures, have been performing finite ability to adapt uncertainty character of real world plant, such as driving dynamics and its desired control performance. This paper introduces a new concept of neural network neuron. In this matter, exploiting discrete z-function builds new neuron activation. Instead of zero order joint weight matrices, the discrete z-function weight matrix will be provided to realize uncertainty or undetermined real word plant and desired adaptive control system that their order has probably been changing. Instead of using bias, an initial condition value is developed. Neural networks using new neurons is called Varied Order Neural Network (VONN). For optimization process, updating order, coefficient and initial value of node activation function uses GA; while updating joint weight, it applies both back propagation (combined LSE-gauss Newton) and NPSO. To estimate the number of hidden layer, constructive back propagation (CBP) was also applied. Thorough simulation was conducted to compare the control performance between FONN and MONN. In order to control, vehicle stability was equipped by electronics stability program (ESP), electronics four wheel steering (4-EWS), and active suspension (AS). 2000, 4000, 6000, 8000 data that are from TODS, a hidden layer, 3 input nodes, 3 output nodes were provided to train and test the network of both the uncertainty model and its adaptive control system. The result of simulation, therefore, shows that stability parameter such as yaw rate error, vehicle side slip error, and rolling angle error produces better performance control in the form of smaller performance index using FDNN than those using MONN.