Search

Real-time information mining of a big dataset consisting of time series data is a very challenging task. For this purpose, we propose using the mean distance and the standard deviation to enhance the accuracy of the existing fast incremental model tree with the drift detection (FIMT-DD) algorithm. The standard FIMT-DD algorithm uses the Hoeffding bound as its splitting criterion. We propose the further use of the mean distance and standard deviation, which are used to split a tree more accurately than the standard method. We verify our proposed method using the large Traffic Demand Dataset, which consists of 4,000,000 instances; Tennet’s big wind power plant dataset, which consists of 435,268 instances; and a road weather dataset, which consists of 30,000,000 instances. The results show that our proposed FIMT-DD algorithm improves the accuracy compared to the standard method and Chernoff bound approach. The measured errors demonstrate that our approach results in a lower Mean Absolute Percentage Error (MAPE) in every stage of learning by approximately 2.49% compared with the Chernoff Bound method and 19.65% compared with the standard method.

Prediction of bone age from an x-ray is one of the methods in the medical field to support predicting endocrine gland disease, growth abnormalities, and genetic disorders. A decision support system to predict the bone age from the x-ray image has been implemented. It utilizes traditional machine learning methods and deep learning. We propose the Region-Based Feature Connected Layer (RB-FCL) from the essential segmented region of hand x-ray. We treat the deep learning models as the feature extraction for each region of the hand x-ray bone. The Feature Connected Layers are the output from the trained important region, such as 1-radius-ulna, 2-carpal, 3-metacarpal, 4-phalanges, and 5-ephypisis. DenseNet121, InceptionV3, and InceptionResNetV2 are the deep learning models that we used to train the critical region. From the evaluation results, the Mean Absolute Error (MAE) results produced is 6.97. This result is better compared to standard deep learning models, which are 9.41.

We propose a Chernoff-bound approach and examine standard deviation value to enhance the accuracy of the existing fast incremental model tree with the drift detection (FIMT-DD) algorithm. It is a data stream mining algorithm that can observe and form a model tree from a large dataset. The standard FIMT-DD algorithm uses the Hoeffding bound for its splitting criterion. The use of the simplified Chernoff bound is proposed for splitting a tree more accurately than the Hoeffding bound. We verify our proposed Chernoff bound and standard deviation value examination by evaluating several large real-world datasets that consists of 100,000,000 instances. The result shows that our proposed method has an improvement in term of accuracy compared to the Hoeffding bound in the FIMT-DD algorithm. Error value measurements demonstrate that the Chernoff bound approach contributes to smaller error value compared to the standard method. In term of overall simulation time, the Chernoff bound approach increases the duration of simulation time and utilizes more memory compared to Hoeffding bound. However, Chernoff bound is much faster to get a smaller error compared to the standard algorithm.

Intelligent transport systems (ITS) are a promising area of studies. One implementation of ITS are advanced driver assistance systems (ADAS), involving the problem of obstacle detection in traffic. This study evaluated the YOLOv4 model as a state-of-the-art CNN-based one-stage detector to recognize traffic obstacles. A new dataset is proposed containing traffic obstacles on Indonesian roads for ADAS to detect traffic obstacles that are unique to Indonesia, such as pedicabs, street vendors, and bus shelters, and are not included in existing datasets. This study established a traffic obstacle dataset containing eleven object classes: cars, buses, trucks, bicycles, motorcycles, pedestrians, pedicabs, trees, bus shelters, traffic signs, and street vendors, with 26,016 labeled instances in 7,789 images. A performance analysis of traffic obstacle detection on Indonesian roads using the dataset created in this study was conducted using the YOLOv4 method.

This paper propose a method for determining the periodicity of dynamic textures from water video using spatial feature similarity measure of Galois-Field (GF) texture representation. Auto correlation function are use to analyze the extracted spatial feature from the representation to determine the periodicity of dynamic textures. There are two type of spatial feature to be compared, the first is histogram and second is normalize cumulative histogram (NCH). Two type of experiment are conducted, the first is virtual rotation where video is rotated virtually from 0o until 360o, the second is actual rotation where camera are rotated physically. Experiments show that although GF improves the performance of the Histogram similarity measure, overall NCH shows better performance. In virtual rotation experiment, GF representation prove to minimize variability due to rotation of camera, the maximum variability produce by NCH is 27%, while when GF are not use the maximum variability is 106%. Contrary, in actual rotation experiment, GF is not proven to minimize variability where NCH produce maximum variability is 57%, while where GF are not use the maximum variability is 9%. The difference in variability pattern between virtual and actual rotation, shows that Galois Field is good at handling dynamic texture rotation, but not against other factors that affect variability.

Traditional food knowledge (TFK) is an essential aspect of human life. In terms of sociocultural aspects, TFK is necessary to protect ancestral culture. In terms of health, traditional foods contain better and more natural ingredients compared to the ingredients of processed foods. Considering this background, in this study, data acquisition and automatic food recognition were performed for traditional food in Indonesia. The food images were captured in a professional mini studio. The food image data were captured under the same light intensity, camera settings, and shooting distance from the camera. The parameters were precisely measured and configured with a light intensity meter, adjustable lighting, and a laser distance measurement device. The data of 1644 traditional food images were successfully obtained in the data acquisition process. These images corresponded to 34 types of traditional foods, and 30–50 images were obtained for each type of food. The size of the raw food image data was 53 GB. The data were divided into sets for training, testing, and validation. An automatic recognition system was developed to classify the traditional food of Indonesia. Training was performed using several types of convolutional neural network (CNN) models such as Densenet121, Resnet50, InceptionV3, and Nasnetmobile. The evaluation results indicated that when using a high quality dataset, the automatic recognition system could realize satisfactory area under the receiver operating characteristics (AUROC) and high accuracy, precision, and recall values of more than 0.95.

Nowadays, the implementation of deep learning in various fields of data mining and big data analytics has been widely used in many applications. This happens because of the ability of deep learning to be able to have excellent performance in terms of predicting cases of classification, features engineering, and clustering. Images data or texts that have large dimensions can also be processed iteratively by deep learning. In the medical field, deep learning is widely used as computer-aided detection (CAD) to provide a decision support system for radiologists or practitioners. In this research, we try to do a deep performance evaluation of several deep learning network models for the Chest X-ray disease decision support system (DSS). These X-ray images are enormous, it consists of 110,120 images, and the size is about 44 GB. Our primary interest is to get a detailed performance profile for every deep learning network model. It consists of a few evaluation aspects, accuracy performance by using areas under the receiver operating characteristic (AUROC), evaluation of training and testing time, investigation of memory usage, observation of central processing unit (CPU) usage, graphics processor power consumption, and provide some improvement solutions. We also offered a few solutions and suggestions to help the doctor or practitioner to select the most effective deep learning network model.

Rapid technology growth has stimulated automated systems development in the medical field. Automatic bone age evaluation is an example of the implementation of this technology in the medical field. The automated assessment is done based on the left-hand X-ray images. This assessment helps the radiologist and pediatrician evaluate children's growth. A system that can generate a precise and reliable prediction is essential. Thus the main challenge is to determine the suitable technology that can generate a reliable forecast, mainly when working with a large quantity of data. Big data is a growing trend; practical computing challenges created by data streams can be found in several types of applications. It is known that data streams are usually obtained from sensors and monitors which accumulatively can make data very large in volume. This can result in the inability of real-time processing to be carried out. In this paper, the data stream technique is utilized to assess and predict bone ages. The analysis process is carried out in real-time when the data arrives so that the process of storing new data is done after the data is analyzed. A 9 GB sized-dataset consisting of 12,611 images were used. The images have various resolutions. We extracted and analyzed image features by using Canny Edge feature extraction. To predict bone age from those extracted features, we enhance the data stream mining technique base on tree stream mining. We use MAE or Mean Absolute Error, RMSE or Root Mean Squared Error, and MAPE or Mean Absolute Percentage Error as metrics in this measurement. The outcomes of our experiment show that our approach in data stream mining can increase performance measurements. The MAPE of our approach gives a 7% lower error evaluation compared to the standard method.

In recent years, automated systems have been implemented in several medical application tasks. For instance, computerized assessment of bone age which utilized by radiologist and pediatrician. This tool is beneficial for monitoring growth condition in children. However, considering the right technique for constructing this computerized bone evaluation is crucial, specifically when enormous data used. In this study, 9 GB X-ray dataset with varied specifications is examined. To address the challenge, we adopt and compare deep learning with several machine learning techniques. ResNet50 which categorized as a deep learning technique is implemented in this study. In contrast, machine learning technique utilizes two keypoints features: Bag of Scale Invariant Feature Transform (SIFT) and Bag of Speeded Up Robust Features (SURF). Bone age is predicted based on those extracted features by several machine learning regression techniques, including Multilayer Perceptron Regressor, Support Vector Regressor, Random Forest Regressor, and XGBoost Regressor. Both techniques are compared quantitatively using three different measurements: Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and Symmetric Mean Absolute Percentage Error (SMAPE). Experiment results indicate that ResNet50 outperforms overall machine learning techniques. ResNet50 achieves 13.58 months of MAE, 17.14 months of RMSE and 16.81% of SMAPE.

In a democratic state, General Election (Pemilu) is a procedure for selecting regional heads regulated in Article 1 paragraph 3 of the 1945 Constitution. KPU (Komisi Pemilihan Umum) is a state institution that organizes by prioritize transparency and accountability in each stage of general elections in Indonesia. One form of openness that has always been in the media spotlight is the vote counting process. The manual calculation process carried out by the General Election Commissions (KPU) on form C1 is time-consuming and resourceful because it involves paid volunteers. In this study, the authors used the proposed method to build a numerical handwriting recognition system on the C1 KPU form. Method proposed is a recognition flow including table detection with candidate contour techniques, feature matching, number segmentation, and digit classification with the convolutional neural network (CNN). The datasets used are from the official KPU election websites in 2014 and 2019. We use capsnet to classify each segmented digit with 95.65% accuracy. The trained model was tested using validation form and reach 80.73% document accuracy using 2019 election form.

Diabetic retinopathy is a disease caused by the complications of diabetes mellitus and can cause blindness. In this study, we classified the stages of diabetic retinopathy using a large-scale dataset that consists of 35,126 fundus images. The classification of diabetic retinopathy includes five stages, from normal to proliferative diabetic retinopathy. In the proposed approach, a morphological feature extraction method and advanced local binary patterns were employed to extract blood vessel and texture features, respectively. The support vector machine, K-nearest neighbor, random forest, and logistic regression techniques were compared as classifiers. The classification was conducted on fundus images from a Kaggle dataset. The experimental results show that the texture feature extraction method based on advanced local binary patterns leads to higher accuracy, precision, and recall score than the blood vessel features extracted using morphological operators.

Information extraction using distributed sensors has been widely used to obtain information knowledge from various regions or areas. Vehicle traffic data extraction is one of the ways to gather information in order to get the traffic condition information. This research intends to predict and visualize the traffic conditions in a particular road region. Traffic data was obtained from Department of Transport UK. These data are collected using hundreds of sensors for 24?h. Thus, the size of data is very huge. In order to get the behavior of the traffic condition, we need to analyze the huge dataset which was obtained from the sensors. The uses of conventional data mining methods are not sufficient to use, due to the process of knowledge building that should store data temporary in the memory. The fact that data is continuously becoming larger over time, therefore we need to find a method that could automatically adapt to process data in the form of streams. We use method called FIMT-DD (Fast Incremental Model Trees-Drift Detection) to analyze and predict the very large traffic dataset. Based on the prediction system that we have developed, we also visualize the prediction of traffic flow condition within generated sensor point in the real map simulation.

The success criteria of a project are the timeliness of completion, the costs incurred not exceeding the budget, the client's needs are met, and the team management which is running well. Some of research found that more than half of the IT projects experience failure. Variables used to determine the most influential factor in software development and implementation success consist of criteria, factor category and factors. Particularly in Verint where the case study took place, there is currently no research on the ranking of factors that affect the success of IT projects. This paper presents quantitative method to perform data analysis using Analytic Hierarchy Process (AHP). The result shows that skilled staff from management attitude category is the most influential factor in software development and implementation project in Verint.

Most of Soil Type in Biak Island, Papua is Coral Limestone. This limestone is used as building material. Limestone mining is one of the income sources for local people. This study tries to map limestone mining sites using multi-sensor remote sensing data fusion and Object-Based Image Analysis (OBIA) classification approach. 1.5 meters resolution SPOT-6 data acquired in 2015 and 2017 used as spectral and geometric parameters in OBIA classification process. Surface deformation points obtained from the PS-InSAR technique on Sentinel-IA SLC SAR data acquired from November 2017 to May 2018 is used as the structural variable for OBIA classification process to determine whether mining site is active or inactive. The overall accuracy of classification result is 84.7% for 2015 SPOT-6 data and 74.9% for 2017 SPOT-6 data.

Traffic congestion is a problem that often occurs in the big cities in Indonesia. It is caused by very rapid increase of vehicle. The offered solution is to monitor the traffic situation automatically. We implemented the method of detecting vehicle during night in four single board computers (SBC) that are: Raspberry Pi B+, Beagleboard Xm, Raspberry Pi 2 and Odroid XU4. Perfomance of Odroid XU4 exceed other single board computers in which the maximum fps obtained 30 frame per second(fps) and the maximum accuracy of vehicle detection reached 98 percent.

In vitro fertilization technology is used to help couples get children. During the development of IVF, embryological will observe the process of cleavage embryo until it is determined which gives the highest probability to produce a pregnancy. During this division process, the observation is done manually by embryological which produce subjective assessment of an embryo and vulnerable reduced quality embryos. Embryo quality is reduced due to the observation carried out outside a developed embryo. In addition to the number of embryos that increasingly divide and have a morphology that are difficult to observe, these judgements are prone to error than the embryological own. This research proposed a method to predict the number of blastameres of the embryo time-lapse using Conditional Random Field (CRF) based on Bag of Visual Words (BoVW). BoVW approach is used to represent data with the purpose of solving the problem of subjectivity embryological votes. The data used for the experiment is the data of the human embryo from the hospital Cipto Mangun Kusumo (RSCM) and data from mouse embryos. Data embryos RSCM has more variations than data in mouse embryo. Based on the experimental results, the use of BoVW able to overcome the problem of subjectivity embryological votes with an accuracy of? 80%. Besides the experimental results with the proposed method uses data embryos RSCM which difficulty level is higher than the data from moose embryos, they were able to identify with both the average accuracy of 96.79%.

Carbon Dioxide gas (CO2) gas contained in our air which has many roles in environment, but in a huge amount it became dangerous. To encounter that, the system for CO2 monitoring is needed. One of the most effective way is using Wireless Sensor Network (WSN). This system capable to monitor concentration of CO2 and another variable using sensor nodes. But not all of that parameters is correlated to concentration of CO2. To make monitoring system runs efficiently, correlation analysis between variables is needed. This research conduct a correlation analysis between concentration of CO2 and humidity, temperature and light intensity from data collected by our own-made digital sensor node. Data gathered for seven days in one location with a fluctuate environment condition. Correlation calculated with Spearman's rho method. The result is CO2 and air humidity have a strong positive correlation with air humidity (0.726), weak negative correlation with light intensity (−0.319), and no correlation with air temperature (−0.008).

This paper proposed a method to build knowledge from one and a half years of UK traffic data sets. The method used is the Fast Incremental Model Trees - Drift Detection (FIMT-DD) with an improvement on the perceptron rule. In order to predict a traditional data set, we first analyze the model. After we have analyzed the model, we then average it from different arrangements of the datasets. In a stream data set, the approach is different from the traditional data sets. The approach of a stream data set is to take several snapshots during the induction to analyze the accuracy progress of the predicted model. We used the tanh activation function to optimize the perceptron rule. Therefore, we are able to reduce the error of the result. The error measurements that we analyzed in this paper are MAE, RMSE, and SMAPE. From 100,000,000 instances of traffic data, the optimized FIMT-DD algorithm method proves to be successful with smaller error value than standard FIMT-DD algorithm. To be able to measure the performance of the error of our optimized FIMT-DD algorithm, we used evaluate prediction sequential to analyze the data. According to the error measurement results of the MAE, RMSE, and SMAPE, tanh(x) activation function has a good influence to decrease the error value and improve the accuracy. The smaller error values did not only happen in a few experiments instances, but it occurs in the majority of the instances evaluation.