Hyeseon Lee

Mining churning behaviors and developing retention strategies based on a partial least squares (PLS) model

In a very competitive mobile telecommunication business environment, marketing managers need a business intelligence model that allows them to maintain an optimal (at least a near optimal) level of churners very effectively and efficiently while minimizing the costs throughout their marketing programs. As a first step toward optimal churn management program for marketing managers, this paper focuses on building an accurate and concise predictive model for the purpose of churn prediction utilizing a partial least squares (PLS)-based methodology on highly correlated data sets among variables. A preliminary experiment demonstrates that the presented model provides more accurate performance than traditional prediction models and identifies key variables to better understand churning behaviors. Further, a set of simple churn marketing programs-device management, overage management, and complaint management strategies-is presented and discussed.

High Diagnostic Accuracy of Antigen Microarray for Sensitive Detection of Hepatitis C Virus Infection

BACKGROUND Hepatitis C virus (HCV) can be transmitted through blood transfusion. Screening ELISA, the most widely used method for HCV diagnosis, sometimes yields false-positive and false-negative results, so a confirmatory test is used. This secondary testing is labor-intensive and expensive, and thus is impractical for massive blood bank screening. Therefore, a new massive screening method with high accuracy is needed for sensitive and specific detection of HCV. METHODS With sol-gel material, we designed novel antigen microarray in 96-well plates for HCV detection. Each individual well was spotted with 4 different HCV antigens. We used this new system to test 154 patient serum samples previously tested for HCV by ELISA (87 HCV positive and 67 HCV negative) (HCV EIA3.0, ABBOTT). We assessed the detection limit of our microarray system with the use of serial 10-fold dilutions of an HCV-positive sample. RESULTS Our microarray assay was reproducible and displayed higher diagnostic accuracy (specificity) (98.78%) than did the ELISA (81.71%). Our method yielded significantly fewer false-positive results than did the ELISA. The detection limit of our assay was 1000 times more sensitive than that of the ELISA. In addition, we found this novel assay technology to be compatible with the currently employed automated methods used for ELISA. CONCLUSION We successfully applied the sol-gel-based protein microarray technology to a screening assay for HCV diagnosis with confirmatory test-level accuracy. This new, inexpensive method will improve the specificity and sensitivity of massive sample diagnosis.

Near-infrared spectral data transfer using independent standardization samples: a case study on the trans-alkylation process

Abstract A variety of standardization or transfer methods between near infrared spectrometric instruments are applied for the content prediction of five major constituents of the product at trans-alkylation process with spectra measured on two different instruments. Because process samples are difficult to be stored, we use independent transfer samples by blending some pure materials for the spectrum standardization of the process samples. Using the independent standardization samples, we investigate the transfer performance of well-known piecewise direct standardization combined with several regression methods on the raw spectra. Also, we propose some indirect standardization methods utilizing wavelet transferred scores or factor scores through principal component analysis and partial least squares. The standardization by transferring scores takes only a few transfer coefficients, but it shows similar performance to the spectrum transfer case. In addition, we show the possibility of using a fewer number of stable samples than the original set of samples for the standardization with similar performance.

A control chart using an auxiliary variable and repetitive sampling for monitoring process mean

In this paper, a new control chart is proposed by using an auxiliary variable and repetitive sampling in order to enhance the performance of detecting a shift in process mean. The product-difference type estimator of the mean is plotted on the proposed control chart, which utilizes the information of an auxiliary variable correlated with the main quality variable. The proposed control chart is based on the outer and inner control limits so that repetitive sampling is allowed when the plotted statistic falls between the two limits. The average run length (ARL) of the proposed control chart is evaluated using the Monte Carlo simulation. The proposed control chart is compared with the Riaz M control chart and the results show the outperformance of the proposed control chart in terms of the ARL.

A variables repetitive group sampling plan under failure-censored reliability tests for Weibull distribution

We propose a variables repetitive group sampling plan under type-II or failure-censored life testing when the lifetime of a part follows a Weibull distribution with a known shape parameter. The acceptance criteria do not involve unknown scale parameter differently from the existing plans. To determine the design parameters of the proposed plan, the usual approach of using two points on the operating characteristic curve is adopted and an optimization problem is formulated so as to minimize the average number of failures observed. Tables for design parameters are constructed when the quality of parts is represented by the unreliability or the ratio of the mean lifetime to the specified life. It is found that the proposed sampling plan can reduce the sample size significantly than do the single sampling plan.

Churn management optimization with controllable marketing variables and associated management costs

Highlights? Formulate churn management problem as an optimization problem while minimizing cost. ? Identify controllable marketing variables and assign management costs for optimization. ? Solve a global churn management problem using partial least square optimization method. ? Select customers based on churn probability or management costs as a local optimization. ? Combine optimization models designed for entire customers and subsets of customers. In this paper, we propose a churn management model based on a partial least square (PLS) optimization method that explicitly considers the management costs of controllable marketing variables for a successful churn management program. A PLS prediction model is first calibrated to estimate the churn probabilities of customers. Then this PLS prediction model is transformed into a control model after relative management costs of controllable marketing variables are estimated through a triangulation method. Finally, a PLS optimization model with marketing objectives and constraints are specified and solved via a sequential quadratic programming method. In our experiments, we observe that while the training and test data sets are dramatically different in terms of churner distributions (50% vs. 1.8%), four controllable variables in three marketing strategies significantly changed through optimization process while other variables only marginally changed. We also observe that the most significant variable in a PLS prediction model does not necessarily change most significantly in our PLS optimization model due to the highest management cost associated, implying differences between a prediction and an optimization model. Finally, two marketing models designed for targeting the subsets of customers based on churn probability or management costs are presented and discussed.

Improving the classification accuracy for IR spectroscopic diagnosis of stomach and colon malignancy using non-linear spectral feature extraction methods

Non-linear feature extraction methods, neighborhood preserving embedding (NPE) and supervised NPE (SNPE), were employed to effectively represent the IR spectral features of stomach and colon biopsy tissues for classification, and improve the classification accuracy for diagnosis of malignancy. The motivation was to utilize the NPE and SNPEs capability of capturing non-linear spectral behaviors by simultaneously preserving local relationships in order that minute spectral differences among classes would be effectively recognized. NPE and SNPE derive an optimal embedding feature such that the local neighborhood structure can be preserved in reduced spaces (variables). The IR spectra collected from stomach and colon tissues were represented by several new variables through NPE and SNPE, and also by using the principal component analysis (PCA). Then, the feature-extracted variables were subsequently classified into normal, adenoma and cancer tissues by using both k-nearest neighbor (k-NN) and support vector machine (SVM), and the resulting accuracies were compared with each other. In both cases, the combination of SNPE-SVM provided the best classification performance, and the accuracy was substantially improved compared to when PCA-SVM was used. Overall results demonstrate that NPE and SNPE could be potential feature-representation strategies useful in biomedical diagnosis based on vibrational spectroscopy where effective recognition of minute spectral differences is critical.

A Variable Selection Procedure for X-ray Diffraction Phase Analysis

The X-ray diffraction method has been widely used for qualitative and quantitative phase abundance analysis of crystalline materials. We propose the use of partial least squares when building the calibration model for a quantitative phase analysis based on X-ray diffraction spectra. We also propose a variable selection procedure to reduce the measurement points in terms of angles as an alternative to using the whole pattern. The proposed method is based on the variable importance in projection derived from the partial least squares and it considers some practical issues regarding the angle measurement. The method was particularly applied to the simultaneous determination of weight fractions of some iron oxides. It was found that the number of measurement points can be reduced to 30 percent of the total number of points with a small sacrifice in prediction error.

Rapid determination of FeO content in sinter ores using DRIFT spectra and multivariate calibrations

Abstract Using diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS), several multivariate calibration methods are explored for the quantitative determination of FeO content in sinter ores. The multivariate calibrations include ridge regression with variable selection, principal component regression, ridge principal component regression, and partial least square regression with the linear and the nonlinear mapping using neural networks. Spectral data are preprocessed by signal correction and scaling prior to the modeling. Cross validation is employed to obtain the optimal biasing parameter in ridge-related regression and to obtain the optimal number of principal components (or latent variables) in component-related modeling. We consider the possibility of reducing the number of variables involved in models while maintaining the prediction power to propose a final prediction model. For the quantitative determination of FeO content in sinter ores, component related regressions on auto-scaled orthogonal signal correction are suggested as appropriate calibration methods.

Haptic Assistance for Memorization of 2-D Selection Sequences

This paper investigates the effect of haptic feedback on the learning of a 2-D sequential selection task, used as an abstraction of complex industrial manual assembly tasks. This mnemonic-motor task requires memorization of the selection order of points scattered on a 2-D plane and reproduction of this order using entire arm movements. Four information presentation methods, visual information only, visual information + enactment, visual information + haptic guidance, and visual information + haptic disturbance, are considered. The latter three methods provide different levels of haptic kinesthetic feedback to the trainee. We carried out a user study to assess the quantitative performance differences of the four training methods using a custom-built visuo-haptic training system. Experimental results showed the relative advantages and disadvantages of each information presentation method for both short-term and long-term memorization. In particular, training with only visual information was the best option for short-term memory, while training also with haptic disturbance was the most effective for long-term memory. Our findings have implications to designing a training method that is suitable for given training requirements.