Bokyoung Kang

Real-time business process monitoring method for prediction of abnormal termination using KNNI-based LOF prediction

In this paper, we propose a novel approach to real-time business process monitoring for prediction of abnormal termination. Existing real-time monitoring approaches are difficult to use proactively, owing to unobserved data from gradual process executions. To improve the utility and effectiveness of real-time monitoring, we derived a KNNI (k nearest neighbor imputation)-based LOF (local outlier factor) prediction algorithm. In each monitoring period of an ongoing process instance, the proposed algorithm estimates the distribution of LOF values and the probability of abnormal termination when the ongoing instance is terminated, which estimations are conducted periodically over entire periods. Thereby, we can probabilistically predict outcomes based on the current progress. In experiments conducted with an example scenario, we showed that the proposed predictors can reflect real-time progress and provide opportunities for proactive prevention of abnormal termination by means of an early alarm. With the proposed method, abnormal termination of an ongoing instance can be predicted, before its actual occurrence, enabling process managers to obtain insights into real-time progress and undertake proactive prevention of probable risks, rather than merely reactive correction of risk eventualities.

Periodic performance prediction for real‐time business process monitoring

Purpose – The purpose of this paper is to provide industrial managers with insight into the real‐time progress of running processes. The authors formulated a periodic performance prediction algorithm for use in a proposed novel approach to real‐time business process monitoring.Design/methodology/approach – In the course of process executions, the final performance is predicted probabilistically based on partial information. Imputation method is used to generate probable progresses of ongoing process and Support Vector Machine classifies the performances of them. These procedures are periodically iterated along with the real‐time progress in order to describe the ongoing status.Findings – The proposed approach can describe the ongoing status as the probability that the process will be executed continually and terminated as the identical result. Furthermore, before the actual occurrence, a proactive warning can be provided for implicit notification of eventualities if the probability of occurrence of the gi...

Fast outlier detection for very large log data

Density-based outlier detection identifies an outlying observation with reference to the density of the surrounding space. In spite of the several advantages of density-based outlier detections, its computational complexity remains one of the major barriers to its application. The purpose of the present study is to reduce the computation time of LOF (Local Outlier Factor), a density-based outlier detection algorithm. The proposed method incorporates kd-tree indexing and an approximated k-nearest neighbors search algorithm (ANN). Theoretical analysis on the approximation of nearest neighbor search was conducted. A set of experiments was conducted to examine the performance of the proposed algorithm. The results show that the method can effectively detect local outliers in a reduced computation time.

Real‐time business process monitoring using formal concept analysis

Purpose – The purpose of this paper is to help industrial managers monitor and analyze critical performance indicators in real time during the execution of business processes by proposing a visualization technique using an extended formal concept analysis (FCA). The proposed approach monitors the current progress of ongoing processes and periodically predicts their probable routes and performances.Design/methodology/approach – FCA is utilized to analyze relations among patterns of events in historical process logs, and this method of data analysis visualizes the relations in a concept lattice. To apply FCA to real‐time business process monitoring, the authors extended the conventional concept lattice into a reachability lattice, which enables managers to recognize reachable patterns of events in specific instances of business processes.Findings – By using a reachability lattice, expected values of a target key performance indicator are predicted and traced along with probable outcomes. Analysis is conduct...

Real-time Process Quality Control for Business Activity Monitoring

This paper proposes a real-time process quality control methodology for Business Activity Monitoring by estimating the risk level (RL) and its confidence intervals (CIs) of an ongoing process. We formulate algorithms to estimate RL and CIs as a continuous measure, based on the partial information of the ongoing process. By using the proposed method, RL and CIs can be estimated in real-time during process execution, so as to effectively control the process quality. Furthermore, it is feasible to generate early warnings based on the status of RL and CIs. We also provided experiments to compare the proposed method with existing approaches, which showed that the proposed method can detect the risks of the ongoing process more precisely and even earlier than existing approaches.

Performance comparison of classifiers for differentiation among obstructive lung diseases based on features of texture analysis at HRCT

The performance of classification algorithms for differentiating among obstructive lung diseases based on features from texture analysis using HRCT (High Resolution Computerized Tomography) images was compared. HRCT can provide accurate information for the detection of various obstructive lung diseases, including centrilobular emphysema, panlobular emphysema and bronchiolitis obliterans. Features on HRCT images can be subtle, however, particularly in the early stages of disease, and image-based diagnosis is subject to inter-observer variation. To automate the diagnosis and improve the accuracy, we compared three types of automated classification systems, naïve Bayesian classifier, ANN (Artificial Neural Net) and SVM (Support Vector Machine), based on their ability to differentiate among normal lung and three types of obstructive lung diseases. To assess the performance and cross-validation of these three classifiers, 5 folding methods with 5 randomly chosen groups were used. For a more robust result, each validation was repeated 100 times. SVM showed the best performance, with 86.5% overall sensitivity, significantly different from the other classifiers (one way ANOVA, p<0.01). We address the characteristics of each classifier affecting performance and the issue of which classifier is the most suitable for clinical applications, and propose an appropriate method to choose the best classifier and determine its optimal parameters for optimal disease discrimination. These results can be applied to classifiers for differentiation of other diseases.

Online process monitoring scheme for fault detection based on Independent Component Analysis (ICA) and Local Outlier Factor (LOF)

Process monitoring in manufacturing field such as chemistry and pharmacy is valuable to reduce the risk of the accidents and to improve the efficiency of the process. Due to the increased size and complicated technology of the modern plants, computer-aided process monitoring has been introduced. Nevertheless, the existing methods for computer-aided process monitoring such as Multivariate Statistical Process Control (MSPC) based on Principal Component Analysis (PCA) or Independent Component Analysis (ICA) present several limits in accurate fault detection of process. To overcome these limits, we introduce a novel online computer-aided process monitoring scheme based on ICA with Local Outlier Factor (LOF). The proposed scheme adopts LOF as the monitoring statistic. By utilizing LOF, the limits of existing related works using MSPC were successfully resolved. The scheme was tested by widely used benchmark dataset of Tennessee Eastman process. The results showed the effectiveness of the proposed scheme.

Agent Based Modeling and Simulation of Structural Hole Based Order Allocation Strategy

Order allocation is one of the most important decision-making problems of firms having significant influences on performances of themselves and the whole supply chain. Existing researches about order allocation have mainly focused on evaluating capabilities of directly connected suppliers so that it is hard to consider effects and interactions from undirected connections over multiple lower-layers. To alleviate the limitation, this paper proposed a novel approach to order allocation using structural hole. By applying the concept of structural hole to the supply network, we could evaluate the structural supplying powers of firms with respect to both of direct and indirect connections. In the proposed approach, we derived a methodology to measure the potential supplying power of each firm by modifying the effective size as one of the measurements of structural hole and then, proposed its application, the structural hole based order allocation strategy. Furthermore, we conducted the agent based modeling of supply chain to perform the decision-making process of order allocation and simulated the proposed strategy. As a results, by coping with the variance of demand more stably, it could improve the performance of supply chain from the aspects of fill rate, inventory level and demand-supply balance.

Real-Time connection monitoring of ubiquitous networks for intrusion prediction: a sequential KNN voting approach

In the ubiquitous network environment where numerous devices are connecting each other, it is believed that security will play an important role in overall network management. And the wireless sensor network (WSN) is commonly considered to be one of such networks prone to a wide range of attacks due to its inherent characteristics. For the sound operation of WSN, it is important to block malicious connections from the network as early as possible. This paper proposes a novel approach to real-time monitoring of network by using the sequential KNN voting. When connection data is sequentially recorded on the log, the final result of ongoing behavior is predicted probabilistically with only partial data, which iterates consecutively as additional connection data are accumulated to the log. Once this predicted probability reaches certain preset threshold value for possible network intrusion, then we can do some preventive actions for this ongoing connection. The value of this research lies in that the eventualities are predicted at the early stage of connection with partial information available. Since the prediction uses sequential KNN voting, the accuracy of our approach can be even more enhanced as with the volume of log grows.