Seungwoo Jeon

Localization of Pallets Based on Passive RFID Tags

As logistics industry is being grown, the saving of logistic operating cost becomes the key of efficiency in many fields of industries. The saving of cost is possible by improvement of processing mechanism efficiently. Simply, if the locations of stored freights are managed on a real time by a certain system, then the cost to find stored freight such as time and fuel of driving forklift is able to be saved. Therefore managing the location of stored freights is important approach to saving logistics operating cost. Therefore, this paper proposes logistics warehouse based on RFID ceiling tag that means RFID tags are attached to the ceiling of logistics warehouse. In our proposal, we use RFID technology to find the location of freights, not RTLS or GPS. Accordingly, adopting RFID technology which is contactless object recognizing system make it possible to automatically monitor and manage location of stored freight and processing status on a real time at relatively low cost in logistics warehouse environment.

Making a Graph Database from Unstructured Text

From a huge volume of text of emails and SNS, it is required to extract human relations to determine whether or not there exist illegal connections each other. A graph structure becomes very useful for giving better representation of human relations compared with the original plain text. In this paper, we propose a way of constructing graph from a number of texts. To make the graph more concise and compact, it is also required to remove duplication and outliers in the graph. The key point of merging a graph structure is to perform automatic and semi-automatic merging method based on our novel merge-feasibility measurement. To justify our new methods of extracting and merging the graph structure, we describe the implementation and testing of our proposed system.

Pattern graph tracking-based stock price prediction using big data

Stock price forecasting is the most difficult field owing to irregularities. However, because stock prices sometimes show similar patterns and are determined by a variety of factors, we propose determining similar patterns in historical stock data to achieve daily stock prices with high prediction accuracy and potential rules for selecting the main factors that significantly affect the price, while simultaneously considering all factors. This study is intended at suggesting a new complex methodology that finds the optimal historical dataset with similar patterns according to various algorithms for each stock item and provides a more accurate prediction of daily stock price. First, we use a Dynamic Time Warping algorithm to find patterns with the most similar situation adjacent to a current pattern. Second, we select the determinants most affected by the stock price using feature selection based on Stepwise Regression Analysis. Moreover, we generate an artificial neural network model with selected features as training data for predicting the best stock price. Finally, we use JaroWinkler distance with Symbolic Aggregate approXimation (SAX) as a prediction accuracy measure to verify the accuracy of our model. We propose a new methodology combined Dynamic Time Warping, Stepwise Regression Analysis, and Artificial Neural Network.We use JaroWinkler distance with Symbolic Aggregate approXimation (SAX) as prediction accuracy measure to verify our model.We construct a big data processing framework to handle the overall processes using big data open sources.

Big Data Processing for Prediction of Traffic Time Based on Vertical Data Arrangement

To predict future traffic conditions in each road with unique spatiotemporal pattern, it is necessary to analyze the conditions based on historical traffic data and select time series forecasting methods which can be predicting next pattern for each road according to the analyzed results. Our goal is to create a new statistical model and a new system for predictive graphs of traffic times based on big data processing tools. First, we suggest a vertical data arrangement, gathering past traffic times in the same time slot for long-term prediction. Second, we analyze each traffic pattern to select time-series variables because a time-series forecasting method for a location and a time will be selected according to the variables that are available. Third, we suggest a spatiotemporal prediction map, which is a two-dimensional map with time and location. Each element in the map represents a time-series forecasting method and an R-squared value as indicator of prediction accuracy. Finally, we introduce a new system including RHive as a middle point between R and Hadoop clusters for generating predicted data efficiently from big historical data.

Computing traffic congestion degree using SNS-based graph structure

Social networking site (SNS) messages can contain subjective traffic information, including congestion-related expressions such as “bad traffic” or “traffic is crazy”. Moreover, they also contain heterogeneous levels of location information, such as a point (latitude, longitude), a road, or an area name, which complicates the process of collecting related traffic information. This paper aims to use SNS messages for monitoring traffic conditions on a road by computing the traffic congestion degree. The process begins by classifying those SNS messages that are related to a road in terms of location information and constructing an initial graph structure to store each message. Because of the heterogeneous location types, we need to combine the initial graph structures based on their spatial references. We can then measure the subjective congestion by computing an expression score using our rule-based approach.

Bigdata analytics on CCTV images for collecting traffic information

Although CCTV is most frequently used for traffic monitoring, there arise problems as a result of which traffic flow needs to be checked manually. In this paper, we propose a new method for automatically calculating traffic volume and vehicle speed by pattern analysis using pixel data extracted from CCTV Video image. First, vertical and horizontal lines are detected that are designated in a lane, and then pixel data are extracted from each frame of the lines. Second, when vehicles pass the lines, they are detected according to pattern change of Value (Brightness) in pixel data.

Flexible Capturing Application for Enhanced Generation of EPCIS Events

Radio frequency identification (RFID) technology and electronic product code (EPC) technology have been widely used to identify and keep track of physical objects. EPCglobal proposed the EPC network which consists of several components such as application level events (ALE) and EPC information service (EPCIS) to deal with the captured data from different layers. Many studies mostly concentrated on dealing with the RFID tag data in ALE, as well as querying and sharing EPCIS events in EPCIS. However, there is no well-known study on specifying how to generate higher level EPCIS events. The event types and semantic event fields are both uncertain for the capturing application to generate EPCIS events. Therefore, this paper proposes the flexible capturing application (FCA) to solve the problem that the event types and semantic event fields are both uncertain. FCA specifies generation rules about the four EPCIS event types. All the generation rules are matched for the incoming tag data to determine the event types. Event fields are generated with tag data and other sources of data after deciding event types. Thus, FCA can generate EPCIS events arising from supply chain activity. We evaluate our approaches by means of simulation and real experiments. Our experimental results indicate that FCA can be effective in processing EPCIS events data. We conclude with suggestions for future work.

Stock Price Prediction based on Stock Big Data and Pattern Graph Analysis

Stock price prediction is extremely difficult owing to irregularity in stock prices. Because stock price sometimes shows similar patterns and is determined by a variety of factors, we present a novel concept of finding similar patterns in historical stock data for high-accuracy daily stock price prediction with potential rules for simultaneously selecting the main factors that have a significant effect on the stock price. Our objective is to propose a new complex methodology that finds the optimal historical dataset with similar patterns according to various algorithms for each stock item and provides a more accurate prediction of daily stock price. First, we use hierarchical clustering to easily find similar patterns in the layer adjacent to the current pattern according to the hierarchical structure. Second, we select the determinants that are most influenced by the stock price using feature selection. Moreover, we generate an artificial neural network model that provides numerous opportunities for predicting the best stock price. Finally, to verify the validity of our model, we use the root mean square error (RMSE) as a measure of prediction accuracy. The forecasting results show that the proposed model can achieve high prediction accuracy for each stock by using this measure.

TPR∗-tree Performance improvement for big tactical moving objects

Because the number of tactical moving object is tens of thousands and the location information of them should be updated in real time while moving, it must be able to manage a large amount of the target data and management of the future target is also required through the predicted query. We have proposed to manage the tactical moving objects using TPR∗-tree in our previous paper. There are various types of moving objects from fighter to navy vessel in the tactical system, the overlapped area between a MBR with the fighters and another MBR with vessel is much bigger. When inserting new target, this leads to degrades performance that all MBRs with overlapped area should be retrieved. For preventing increases in the overlapped area, we suggest a method to improve TPR∗-tree performance with selective data distinction, especially fast data in this paper.

Analysis of the effect of weather determinants on lodging demands using big data processing

Weather conditions determine the variation in the floating population at tourist sites and, consequently, the number of persons who stay overnight; therefore, it is necessary to select those determinants that are most influenced by the weather. Our aim is to construct a multiple linear regression model and system based on big data processing tools to process large volumes of meteorological and population data with the ultimate goal of supporting predicted accommodation congestion with high prediction accuracy. First, we transform the floating population data into accommodation congestion by estimating the resident population during periods of high demand. Second, we conduct a multiple linear regression analysis with variable removal steps and a residual analysis. Moreover, to verify our regression model, we use two prediction accuracy measures: the mean absolute percentage error and the R-squared value. The final part of the paper describes the construction of a big data processing system that was built to compute the series of prediction operations.