Kwok Leung Tsui

A GPU-accel erated parallel Jaya algorithm for efficiently estimating Li-ion battery model parameters

Abstract A parallel Jaya algorithm implemented on the graphics processing unit (GPU-Jaya) is proposed to estimate parameters of the Li-ion battery model in this paper. Similar to the generic Jaya algorithm (G-Jaya), the GPU-Jaya is free of tuning algorithm-specific parameters. Compared with the G-Jaya algorithm, three main procedures of the GPU-Jaya, the solution update, fitness value computation, and the best/worst solution selection are all computed in parallel on GPU via a compute unified device architecture (CUDA). Two types of memories of CUDA, the global memory and the shared memory are utilized in the execution. The effectiveness of the proposed GPU-Jaya algorithm in estimating model parameters of two Li-ion batteries is validated via real experiments while its high efficiency is demonstrated by comparing with the G-Jaya and other considered benchmarking algorithms. The experimental results reflect that the GPU-Jaya algorithm can accurately estimate battery model parameters while tremendously reduce the execution time using both entry-level and professional GPUs.

Detecting node propensity changes in the dynamic degree corrected stochastic block model

Abstract Many applications involve dynamic networks for which a sequence of snapshots of network structure is available over time. Studying the evolution of node propensity over time can be important in exploring and analyzing these networks. In this paper, we propose a multivariate surveillance plan to monitor node propensity in the dynamic degree corrected stochastic block model. The method is flexible enough to detect anomalous nodes that arise from different mechanisms, including individual change, individuals switch, and global change. Experiments on simulated and case study social network data streams demonstrate that our surveillance strategy can efficiently detect node propensity changes in dynamic networks.

An integrated approach for surgery scheduling under uncertainty

Abstract Operating rooms (ORs) account for high costs in hospitals. A well-designed surgery scheduling system can help improve facility utilization, thus reduce the cost. This paper is concerned with a surgery scheduling problem in the context where the number of surgeries in waiting list is beyond the capacity of OR. A surgeon may perform more than one surgery a day, and the surgeries of a surgeon are scheduled consecutively, which form a block. A model is proposed to determine which surgeries should be performed in the coming workday, as well as the corresponding start time of each block. We propose an integrated approach by combining two existing methods, i.e., sample average approximation (SAA) and robust linear programming. The new approach eliminates the number of variables in SAA model, hence can be solved more efficiently. Experiments show that the computation time of our approach is approximately one quarter of that of SAA. Cost sensitivity analysis is provided.

EEMD-Based Steady-State Indexes and Their Applications to Condition Monitoring and Fault Diagnosis of Railway Axle Bearings

Railway axle bearings are one of the most important components used in vehicles and their failures probably result in unexpected accidents and economic losses. To realize a condition monitoring and fault diagnosis scheme of railway axle bearings, three dimensionless steadiness indexes in a time domain, a frequency domain, and a shape domain are respectively proposed to measure the steady states of bearing vibration signals. Firstly, vibration data collected from some designed experiments are pre-processed by using ensemble empirical mode decomposition (EEMD). Then, the coefficient of variation is introduced to construct two steady-state indexes from pre-processed vibration data in a time domain and a frequency domain, respectively. A shape function is used to construct a steady-state index in a shape domain. At last, to distinguish normal and abnormal bearing health states, some guideline thresholds are proposed. Further, to identify axle bearings with outer race defects, a pin roller defect, a cage defect, and coupling defects, the boundaries of all steadiness indexes are experimentally established. Experimental results showed that the proposed condition monitoring and fault diagnosis scheme is effective in identifying different bearing health conditions.

A profile monitoring of the multi-stage process

Abstract The effective monitoring of the multi-stage process in the modern manufacturing and service processes is crucial in maintaining and improving the final output quality. The essential of the multi-stage process monitoring is to be able to give a signal at each single stage in order to avoid the delay in detecting assignable causes in the process. A multi-stage process is separated to a series of the single-stage processes, and each single-stage process is treated as a profile structure, whose quality is considered as the process output with inputs from the previous stage and the current stage. The EWMA chart is implemented to the profile process using the orthogonal design input. The true process output is often measured with errors because the complexity of the multi-stage process makes the true quality difficult to measure. The harmfulness of measurement errors in the multi-stage process is also investigated.

An integrated machine learning framework for hospital readmission prediction

Abstract Unplanned readmission (re-hospitalization) is the main source of cost for healthcare systems and is normally considered as an indicator of healthcare quality and hospital performance. Poor understanding of the relative importance of predictors and limited capacity of traditional statistical models challenge the development of accurate predictive models for readmission. This study aims to develop a robust and accurate risk prediction framework for hospital readmission, by combining feature selection algorithms and machine learning models. With regard to feature selection, an enhanced version of multi-objective bare-bones particle swarm optimization (EMOBPSO) is developed as the principal search strategy, and a new mutual information-based criterion is proposed to efficiently estimate feature relevancy and redundancy. A greedy local search strategy (GLS) is developed and merged into EMOBPSO to control the final feature subset size as desired. For the modeling process, manifold machine learning models, such as support vector machine, random forest, and deep neural network, are trained with preprocessed datasets and corresponding feature subsets. In the case study, the proposed methodology is applied to an actual hospital located in Northeast China, with various levels of data collected from the hospital information system. Results obtained from comparative experiments demonstrate the effectiveness of EMOBPSO and EMOBPSO-GLS feature selection algorithms. The combination of EMOBPSO (EMOBPSO-GLS) and deep neural network possesses robust predictive power among different datasets. Furthermore, insightful implications are abstracted from the obtained elite features and can be used by practitioners to determine the vulnerable patients for readmission and target the delivery of early resource-intensive interventions.

A Framework of Rebalancing Imbalanced Healthcare Data for Rare Events’ Classification: A Case of Look-Alike Sound-Alike Mix-Up Incident Detection

Identifying rare but significant healthcare events in massive unstructured datasets has become a common task in healthcare data analytics. However, imbalanced class distribution in many practical datasets greatly hampers the detection of rare events, as most classification methods implicitly assume an equal occurrence of classes and are designed to maximize the overall classification accuracy. In this study, we develop a framework for learning healthcare data with imbalanced distribution via incorporating different rebalancing strategies. The evaluation results showed that the developed framework can significantly improve the detection accuracy of medical incidents due to look-alike sound-alike (LASA) mix-ups. Specifically, logistic regression combined with the synthetic minority oversampling technique (SMOTE) produces the best detection results, with a significant 45.3% increase in recall (recall = 75.7%) compared with pure logistic regression (recall = 52.1%).

Influence of Flavors on the Propagation of E-Cigarette–Related Information: Social Media Study

Background Modeling the influence of e-cigarette flavors on information propagation could provide quantitative policy decision support concerning smoking initiation and contagion, as well as e-cigarette regulations. Objective The objective of this study was to characterize the influence of flavors on e-cigarette–related information propagation on social media. Methods We collected a comprehensive dataset of e-cigarette–related discussions from public Pages on Facebook. We identified 11 categories of flavors based on commonly used categorizations. Each post’s frequency of being shared served as a proxy measure of information propagation. We evaluated a set of regression models and chose the hurdle negative binomial model to characterize the influence of different flavors and nonflavor control variables on e-cigarette–related information propagation. Results We found that 5 flavors (sweet, dessert & bakery, fruits, herbs & spices, and tobacco) had significantly negative influences on e-cigarette–related information propagation, indicating the users’ tendency not to share posts related to these flavors. We did not find a positive significance of any flavors, which is contradictory to previous research. In addition, we found that a set of nonflavor–related factors were associated with information propagation. Conclusions Mentions of flavors in posts did not enhance the popularity of e-cigarette–related information. Certain flavors could even have reduced the popularity of information, indicating users’ lack of interest in flavors. Promoting e-cigarette–related information with mention of flavors is not an effective marketing approach. This study implies the potential concern of users about flavorings and suggests a need to regulate the use of flavorings in e-cigarettes.

Two-stage aggregation paradigm for HFLTS possibility distributions: A hierarchical clustering perspective

Abstract The integration of possibility distribution into hesitant fuzzy linguistic term set (HFLTS) adds an extra dimension to individual opinion approximation process and significantly leads to enhanced data quality and reliability. However, aggregation of HFLTS possibility distributions involves merely associated possibilities of linguistic terms without taking into account all possible combinations of individual linguistic opinions. Therefore, computing with HFLTS possibility distributions in such a way has a high possibility of distorting final decisions due to loss of information. The introduction of hesitant 2-tuple linguistic term set (H2TLTS), which technically includes the HFLTS as a special case, offers us a different point of view in consolidating the aggregation process of HFLTSs. Due to the resemblance with H2TLTS, the alternative explantation of HFLTS, i.e., possibility distribution, can be analogously adapted to the theory of H2TLTS. By means of a conceptually simple recasting of HFLTS possibility distribution into a unified framework for H2TLTS possibility distribution with the development of possibilistic 2-tuple linguistic pair (P2TLP) concept, we develop a novel two-stage aggregation paradigm for HFLTS possibility distributions. At the first stage, the initial aggregation takes all possible combinations of P2TLPs in separate HFLTS possibility distributions together to generate an aggregated set of P2TLPs. Building on that, the subsequent stage proposes a similarity measure-based agglomerative hierarchical clustering (SM-AggHC) algorithm to reduce the cardinality of the aggregate set under consideration. The centroid approach combined with the normalization process finally guarantees the aggregation outcomes to be operated as H2TLTS possibility distributions.

Optimal designs for dose–response models with linear effects of covariates

Personalized medicine is becoming more and more important nowadays since the efficacy of a certain medicine vary among different patients. This requires to combine the effects of the prognostic factors or covariates along with different dosages when planning a dose–response experiment. Statistically, this corresponds to the construction of optimal designs for estimating dose–response curves in the presence of covariates. Some characteristics of the optimal designs are derived in order to search such optimal designs efficiently, and an equivalence theorem of the locally ϕs-optimal designs is established accordingly. Computational issues are also studied and presented with theoretical backups. As applications of the above theories, the locally optimal designs are searched out in several situations. Some simulations reveal that the searched locally optimal designs are robust to the moderate misspecification of the prespecified parameters.