Li-Fei Chen

Particle swarm optimization for feature selection with application in obstructive sleep apnea diagnosis

Feature selection is a preprocessing step of data mining, in which a subset of relevant features is selected for building models. Searching for an optimal feature subset from a high-dimensional feature space is an NP-complete problem; hence, traditional optimization algorithms are inefficient in solving large-scale feature selection problems. Therefore, meta-heuristic algorithms are extensively adopted to effectively address feature selection problems. In this paper, we propose an analytical approach by integrating particle swarm optimization (PSO) and the 1-NN method. The data sets collected from UCI machine learning databases were used to evaluate the effectiveness of the proposed approach. Implementation results show that the classification accuracy of the proposed approach is significantly better than those of BPNN, LR, SVM, and C4.5. Furthermore, the proposed approach was applied to an actual case on the diagnosis of obstructive sleep apnea (OSA). After implementation, we conclude that our proposed method can help identify important factors and provide a feasible model for diagnosing medical disease.

A Neural-Network Approach for Defect Recognition in TFT-LCD Photolithography Process

Since the advent of high qualification and tiny technology, yield control in the photolithography process has played an important role in the manufacture of thin-film transistor-liquid crystal displays (TFT-LCDs). Through an auto optic inspection (AOI), defect points from the panels are collected, and the defect images are generated after the photolithography process. The defect images are usually identified by experienced engineers or operators. Evidently, human identification may produce potential misjudgments and cause time loss. This study therefore proposes a neural-network approach for defect recognition in the TFT-LCD photolithography process. There were four neural-network methods adopted for this purpose, namely, backpropagation, radial basis function, learning vector quantization 1, and learning vector quantization 2. A comparison of the performance of these four types of neural-networks was illustrated. The results showed that the proposed approach can effectively recognize the defect images in the photolithography process.

Data Mining Techniques for Assisting the Diagnosis of Pressure Ulcer Development in Surgical Patients

Pressure ulcer is a serious problem during patient care processes. The high risk factors in the development of pressure ulcer remain unclear during long surgery. Moreover, past preventive policies are hard to implement in a busy operation room. The objective of this study is to use data mining techniques to construct the prediction model for pressure ulcers. Four data mining techniques, namely, Mahalanobis Taguchi System (MTS), Support Vector Machines (SVMs), decision tree (DT), and logistic regression (LR), are used to select the important attributes from the data to predict the incidence of pressure ulcers. Measurements of sensitivity, specificity, F1, and g-means were used to compare the performance of four classifiers on the pressure ulcer data set. The results show that data mining techniques obtain good results in predicting the incidence of pressure ulcer. We can conclude that data mining techniques can help identify the important factors and provide a feasible model to predict pressure ulcer development.

An improved particle swarm optimization for feature selection

Searching for an optimal feature subset in a high-dimensional feature space is an NP-complete problem; hence, traditional optimization algorithms are inefficient when solving large-scale feature selection problems. Therefore, meta-heuristic algorithms have been extensively adopted to solve the feature selection problem efficiently. This study proposes an improved particle swarm optimization IPSO algorithm using the opposite sign test OST. The test increases population diversity in the PSO mechanism, and avoids local optimal trapping by improving the jump ability of flying particles. Data sets collected from UCI machine learning databases are used to evaluate the effectiveness of the proposed approach. Classification accuracy is employed as a criterion to evaluate classifier performance. Results show that the proposed approach outperforms both genetic algorithms and sequential search algorithms.

Application of Six Sigma Methodology to Optimize the Performance of the Inter-Metal Dielectric Process

As semiconductor device density increases, integrated circuits generally involve more levels of metallization. In multilevel interconnected metallization schemes, an inter-metal dielectric (IMD) is deposited between metal layers. The basic advantage of this dielectric layer is that it provides good step coverage to help smooth the topology, making it free of pinholes and allowing it to act as a good insulator. The key problem in the IMD layer is the occurrence of voids which may cause electric leakage and later result in yield loss. The occurrence of these voids may be avoided by ensuring an excellent gap-fill capacity. Taking this into consideration, improving the gap-fill capacity is one of the critical issues for the IMD layer. In addition, the quantity of fluorine and the value of voltage ramping to dielectrics breakdown (VRDB) also affect the quality of the IMD layer. The gap-fill capacity, VRDB, and the quantity of fluorine are important quality characteristics for the optimization of the performance of the IMD layer. Due to the complicated inputs/responses relation, it is difficult for the output of the IMD process to reach the desired target. To resolve this problem, this study employs the Six Sigma methodology to reduce the defect in the IMD layer. A case study of a semiconductor manufacturing foundry in Taiwan is illustrated to show the practicability of the Six Sigma methodology.

Prediagnosis of Obstructive Sleep Apnea via Multiclass MTS

Obstructive sleep apnea (OSA) has become an important public health concern. Polysomnography (PSG) is traditionally considered an established and effective diagnostic tool providing information on the severity of OSA and the degree of sleep fragmentation. However, the numerous steps in the PSG test to diagnose OSA are costly and time consuming. This study aimed to apply the multiclass Mahalanobis-Taguchi system (MMTS) based on anthropometric information and questionnaire data to predict OSA. Implementation results showed that MMTS had an accuracy of 84.38% on the OSA prediction and achieved better performance compared to other approaches such as logistic regression, neural networks, support vector machine, C4.5 decision tree, and rough set. Therefore, MMTS can assist doctors in prediagnosis of OSA before running the PSG test, thereby enabling the more effective use of medical resources.

A new particle swarm feature selection method for classification

Searching for an optimal feature subset from a high-dimensional feature space is an NP-complete problem; hence, traditional optimization algorithms are inefficient when solving large-scale feature selection problems. Therefore, meta-heuristic algorithms are extensively adopted to solve such problems efficiently. This study proposes a regression-based particle swarm optimization for feature selection problem. The proposed algorithm can increase population diversity and avoid local optimal trapping by improving the jump ability of flying particles. The data sets collected from UCI machine learning databases are used to evaluate the effectiveness of the proposed approach. Classification accuracy is used as a criterion to evaluate classifier performance. Results show that our proposed approach outperforms both genetic algorithms and sequential search algorithms.

Location selection using fuzzy-connective-based aggregation networks: a case study of the food and beverage chain industry in Taiwan

Location selection is a significant part of strategic management activities. The plurality and interactions of the evaluation criteria involved in the problem complicates the decision-making process. This study successfully proposes a set of systematic procedure for location selection using fuzzy-connective-based aggregation networks. The fuzzy-connective-based aggregation network can aggregate the relative status or achievement among locations in various location-related variables through a hierarchical decision-making structure. Finally, an overall evaluation of location is produced from various aspects. The trained model approximates the relationship between turnover and individual location-related factors, which can be used to predict the potential performance of candidate sites and to answer “what if” questions. The transparency and interpretation ability also makes the proposed method desirable. The weights and parameters in the evaluation model help identify the major factors influencing the turnover and the compensatory relationship among location-related factors. The effectiveness and applicability are confirmed through a case study of the food and beverage chain industry in Taiwan.

An improved negative selection approach for anomaly detection: with applications in medical diagnosis and quality inspection

Negative selection (NS) is one of the most discussed algorithms in artificial immune system (AIS). With its unique property for anomaly detection, it has attracted the attention of researchers in the past decades. However, the processes on how to generate representative detectors and how to define the matching rules remain to be challenges in many NS applications. These difficulties make NS suffer from high false-positive rates and computational complexities. On the other hand, the Mahalanobis distance (MD) is a popular distance metric used in distinguishing patterns of a certain group from those of another group. Compared with other multivariate measurement techniques, MD is superior in its ability to determine the similarity of a set of values from an unknown sample to a set of values measured from a collection of known samples. In this study, an MD-based NS called MDNS is proposed to improve the classification power for anomaly detection by providing the mechanism to judge the quality of detector cells as well as to be applied to define the matching rules and the threshold in a matching rule. Two real cases concerning medical diagnosis and quality inspection in highly reliable products are studied, and the results show that the performance of the NS can be significantly improved by using the proposed approach.

A novel framework for customer complaint management

ABSTRACT Complaining is one option available to customers to express their dissatisfaction with inadequate services. Their complaints contain valuable information for service providers to improve customer relationships and operational quality, which can ultimately enhance business profitability. Customer complaints are frequently handled at the individual level, however, which addresses the symptoms rather than the causes of customer dissatisfaction. This paper presents a framework integrating a decision tree approach, a common data mining tool, into Six Sigma methodology to analyze customer complaints in aggregate and improve service quality by identifying and addressing the underlying causes of failed service. A case study of a restaurant chain was used to demonstrate the effectiveness of the proposed framework. The results indicated a significant (60%) decrease in the number of customer complaints received. Subsequent long-term benefits can be expected.