Conventional Regression Techniques Research Articles

Rapid online adaptation based on transformation space model evolution

This paper presents a new approach to online linear regression adaptation of continuous density hidden Markov models based on transformation space model (TSM) evolution. The TSM which characterizes the a priori knowledge of the training speakers associated with maximum likelihood linear regression matrix parameters is effectively described in terms of the latent variable models such as the factor analysis or probabilistic principal component analysis. The TSM provides various sources of information such as the correlation information, the prior distribution, and the prior knowledge of the regression parameters that are very useful for rapid adaptation. The quasi-Bayes estimation algorithm is formulated to incrementally update the hyperparameters of the TSM and regression matrices simultaneously. The proposed TSM evolution is a general framework with batch TSM adaptation as a special case. Experiments on supervised speaker adaptation demonstrate that the proposed approach is more effective compared with the conventional quasi-Bayes linear regression technique when a small amount of adaptation data is available.

Age assessment at the time of death based on panoramic radiography

The determination of age at the time of death is an important method in forensic anthropology and paleodemography. The possible postmortem investigation of the teeth and jaws enables the determination of age at the time of death, as the bones and teeth are both resistant to degradation in soil and characterized by age-related morphological changes. The aim of this study was to determine whether the age-related changes visible on panoramic radiography correlated with age, and enabled the assessment of individual age. Seven radiographic parameters were used in the study: tooth loss, occlusal tooth wear, pulp stones, carious teeth, periapical disease, tooth restoration, and alveolar bone loss associated with periodontal disease. The material comprised 314 dental panoramic tomograms of living patients of both sexes with documented age (18 to 77 years). Multiple regression equations were constructed for the age estimation, including four parameters (the number of missing teeth, the number of intact teeth, the distance of cement-enamel junction from the alveolar ridge, the number of abraded teeth). The nature of data treated by regression analysis required the careful choice of parameters, appropriate functional model for each parameter, and the experience of the investigator. With the four parameters included in equations, the error was +/- 2.55 years. This preliminary analysis showed that the conventional regression technique could be appropriate for the age estimation based on panoramic radiography, and that an additional study with a larger sample and on wider population was required.

Correlations between chemical and age domains in monazite, and metamorphic reactions involving major pelitic phases: an integration of ID-TIMS and SHRIMP geochronology with Y–Th–U X-ray mapping

Chemical mapping and in situ U–Th–Pb analyses reveal a link between age domains and zones of relative yttrium (Y) depletion or enrichment within monazite crystals and are correlated with metamorphic reactions involving garnet. Conventional, small-fraction isotope dilution thermal ionization mass spectrometry (ID-TIMS) and sensitive high-resolution ion microprobe (SHRIMP) techniques were utilized to measure U–Th–Pb isotopic compositions in metamorphic monazite from pelitic rocks of the southern Canadian Cordillera. Monazite ID-TIMS data from individual samples commonly demonstrate a 2 to 25 Ma range in U-Pb ages. This is difficult to reconcile using conventional regression techniques due to complexities such as excess 206Pb or bulk mixing of discrete age domains. Backscattered electron (BSE) imaging and X-ray elemental mapping for Y, Th, and U revealed complex internal zonation within many of the monazite crystals and served as a guide for detailed, in situ (∼30 μm) isotopic analysis using the SHRIMP. The Y maps generally provided the clearest indication of growth or recrystallization domains and were critical for targeting SHRIMP analysis because these relationships were not always clear in BSE, U, and Th images. Moreover, the Y domains consistently correlated with distinct age domains, with up to three or more in some crystals. These data clearly illustrate the cause of age dispersion within the analyzed monazite grains and demonstrate the significance of multiple age domains in metamorphic monazite that may be irreconcilable or misinterpreted when using conventional dating techniques such as ID-TIMS. Recent studies have investigated the interaction between accessory minerals such as monazite and major pelitic phases throughout a metamorphic event. Researchers have begun to focus on the partitioning of Y between monazite and garnet because it is highly compatible in both phases. Due to its greater volume, garnet exerts considerable control over the Y budget available during metamorphism in pelitic rocks. Consequently, this is reflected in the production and consumption of monazite, as it is sensitive to the availability of Y, and is preserved as internal zones of relative Y enrichment or depletion. Thus, precise ages of contrasting Y domains within monazite provided by in situ ion probe analysis may be correlated with metamorphic reactions involving garnet and assigned to points along the P– T path.

A spatial regression procedure for evaluating the relationship between AVHRR-NDVI and climate in the northern Great Plains

The relationship between vegetation and climate in the grassland and cropland of the northern US Great Plains was investigated with Normalized Difference Vegetation Index (NDVI) (1989–1993) images derived from the Advanced Very High Resolution Radiometer (AVHRR), and climate data from automated weather stations. The relationship was quantified using a spatial regression technique that adjusts for spatial autocorrelation inherent in these data. Conventional regression techniques used frequently in previous studies are not adequate, because they are based on the assumption of independent observations. Six climate variables during the growing season; precipitation, potential evapotranspiration, daily maximum and minimum air temperature, soil temperature, solar irradiation were regressed on NDVI derived from a 10-km weather station buffer. The regression model identified precipitation and potential evapotranspiration as the most significant climatic variables, indicating that the water balance is the most important factor controlling vegetation condition at an annual timescale. The model indicates that 46% and 24% of variation in NDVI is accounted for by climate in grassland and cropland, respectively, indicating that grassland vegetation has a more pronounced response to climate variation than cropland. Other factors contributing to NDVI variation include environmental factors (soil, groundwater and terrain), human manipulation of crops, and sensor variation.

Modeling the formation of chlorination by-products in river waters with different quality

Water chlorination results in formation of a variety of organic compounds, known as chlorination by-products (CBPs), mainly trihalomethanes (THMs) and haloacetic acids (HAAs). Factors affecting their concentrations have been found to be organic matter content of water, pH, temperature, chlorine dose, contact time and bromide concentration, but the mechanisms of their formation are still under investigation. Within this scope, chlorination experiments have been conducted with river waters from Lesvos island, Greece, with different water quality regarding bromide concentration and organic matter content. The factors studied were pH, time and chlorine dose. The determination of CBPs was carried out by gas chromatography techniques. Statistical analysis of the results was focused on the development of multiple regression models for predicting the concentrations of total trihalomethanes and total HAAs based on the use of pH, reaction time and chlorine dose. The developed models, although providing satisfactory estimations of the concentrations of the CBPs, showed lower correlation coefficients than the multiple regression models developed for THMs only during previous study. It seems that the different water quality characteristics of the two river waters in the present study is responsible for this phenomenon. The results indicate that under these conditions the formation of THMs and HAAs in water has a more stochastic character, which is difficult to be described by the conventional regression techniques.

A machine learning approach to the estimation of the liquidus temperature of glass-forming oxide blends

Many properties of glasses and glass-forming liquids of oxide mixtures vary in a relatively simple and regular way with the oxide concentrations. In that respect, the liquidus temperature is an exception, which makes its prediction difficult: the surface to be estimated is fairly complex, so that usual regression methods involve a large number of adjustable parameters. Neural networks, viewed as parameterized non-linear regression functions, were proved to be parsimonious : in order to reach the same prediction accuracy, a neural network requires a smaller number of adjustable parameters than conventional regression techniques such as polynomial regression. We demonstrate this very valuable property on some examples of oxide mixtures involving up to five components. In the latter case, we show that neural networks provide a sizeable improvement over polynomial methods.

FLOOD STAGE FORECASTING WITH SUPPORT VECTOR MACHINES1

ABSTRACT: Machine learning techniques are finding more and more applications in the field of forecasting. A novel regression technique, called Support Vector Machine (SVM), based on the statistical learning theory is explored in this study. SVM is based on the principle of Structural Risk Minimization as opposed to the principle of Empirical Risk Minimization espoused by conventional regression techniques. The flood data at Dhaka, Bangladesh, are used in this study to demonstrate the forecasting capabilities of SVM. The result is compared with that of Artificial Neural Network (ANN) based model for one‐lead day to seven‐lead day forecasting. The improvements in maximum predicted water level errors by SVM over ANN for four‐lead day to seven‐lead day are 9.6 cm, 22.6 cm, 4.9 cm and 15.7 cm, respectively. The result shows that the prediction accuracy of SVM is at least as good as and in some cases (particularly at higher lead days) actually better than that of ANN, yet it offers advantages over many of the limitations of ANN, for example in arriving at ANN's optimal network architecture and choosing useful training set. Thus, SVM appears to be a very promising prediction tool.

Applying reverse regression techniques in earnings–return analyses

Measurement error in unexpected earnings is recognized as a source of bias in examinations of the relation between earnings and returns. Reverse regression procedures are commonly used as a means of coping with this bias. This study examines the properties of reverse regression procedures in multi-interacted variable settings with a specific focus on the earnings response coefficient (ERC) analysis of Collins and Kothari (J. Account. Econom. 11 (1989) 143.). It shows that both conventional reverse regression techniques and novel techniques employed by Collins and Kothari are not robust. It also demonstrates how reverse regression techniques can be successfully employed in such settings using non-interacted-variable designs.

Does the Dog Wag the Tail or the Tail the Dog? Cointegration of Indian Agriculture with Nonagriculture

An oft-repeated refrain in the development literature has been the “neglect” of the agricultural sector vis-à-vis the industrial sector in the development process of the less developed economies. Because infrastructure is crucially linked to both agricultural and industrial development, poor infrastructure development may make it appear as if slow agricultural growth has caused slow industrial growth. Further, in estimating the relation between agriculture and industry, the former should not be assumed to be exogenous; rather, this should first be established. Moreover, given the presence of nonstationarity conventional regression techniques may yield spurious regressions and significance tests. To circumvent these various problems, we study the cointegration of the different sectors of the Indian economy in a multivariate vector autoregression framework.

A correction method for heated length effect in critical heat flux prediction

A new correction method is developed for the effect of the length-to-diameter ( L/D) ratio on critical heat flux (CHF) by applying artificial neural networks and conventional regression techniques to the KAIST CHF data base for water flow in uniformly-heated, vertical round tubes. It consists of two parts: (a) a threshold L/D over which the length effect becomes negligible; and (b) a L/D correction factor for channels with L/D less than the threshold L/D. The proposed correction method is validated with the experimental data in the original database and a new data set obtained from the KAIST intermediate pressure loop. The proposed method will be useful in the following applications: (a) to predict the CHF for short tubes using CHF models which are based on the data for sufficiently long channels; (b) to define the experimental data which can be used for development of local-condition type CHF correlations; and (c) to convert the CHF data from short channels into CHF data for standard long channels for utilization in correlation development.

Arms Transfers and Coups d'État: A Study on Sub-Saharan Africa

Early studies of African coups have largely focused on domestic determinants such as social mobilization, economic conditions, and the characteristics of the military itself. Scant attention has been paid to the political effects of arms transfers on military intervention in Africa. Considering that defense spending may meet the demands of the armed forces, absorb vital social and economic resources needed for development, retard the growth of the civilian domestic product, or strengthen the political position of the military, it is likely that expenditures on arms imports may directly or indirectly affect the likelihood of coups in sub-Saharan African countries. Utilizing theories of arms transfers and coups d'état, a causal model was constructed and data were collected for 35 African nations. An event-count analysis in conjunction with conventional regression techniques was employed for statistical estimation. The findings suggest that arms transfers serve to meet the military's corporate interests and have a long-term direct effect on reducing the likelihood of coups in Africa. However, arms transfers contribute to African regime instability indirectly by enhancing the military's position vis-a-vis civilian institutions in society. Former French colonies have experienced significantly fewer coups than the rest of Africa, but throughout sub-Saharan Africa, countries that experienced an economic downturn or a coup are more likely to have military intervention in politics.

Quantile Curves without Crossing

Conventional regression techniques focus on the conditional averages, but often of more interest are lower or upper conditional quantiles. A more informative description of the relationship among variables can be obtained through regression quantiles (RQ). The percentile curves are usually computed one level at a time. Associated with great flexibility is the embarrassing phenomenon of quantile crossing. We propose a restricted version of regression quantiles (RRQ) that avoids the occurrence of crossing while maintaining sufficient modeling flexibility. RRQ remains in the general framework of the regression quantiles both conceptually and computationally. Because it relates all quantile functions through the conditional median, it also means substantial savings in computation costs when multiple quantiles for high-dimensional data are needed. Two examples in connection with the physical and engineering sciences are presented to demonstrate the usefulness of RRQ curves.

Estimating the Isothermal Compressibility Coefficient of Undersaturated Middle East Crudes Using Neural Networks

The isothermal compressibility coefficients are required in several reservoir engineering applications such as transient fluid flow problems and also in the determination of physical properties of crude oils. Over the years, several correlations to estimate PVT properties have been reported in the literature for different types of hydrocarbon systems. All of these correlations were developed using conventional regression or graphical techniques that may not lead to the highest accuracy. On the other hand, the use of neural networks to develop such correlations can be excellent and reliable for the prediction of these properties. This paper presents an artificial neural network model to predict the isothermal compressibility coefficient of undersaturated crude oils of the Middle East region. The back-propagation algorithm with momentum for error minimization was used in this study. The data set, on which the network was trained, contain 520 experimentally obtained PVT data sets, representing 102 different ...

Elements of an Approach to the Assessment of Random Uncertainty in Transient Measurements

This paper presents an approach to the assessment of random uncertainty in transient measurements based on the modeling of a time series that is representative of the variable of interest. The recommended approach employs conventional regression techniques to estimate all parameters that appear in a model for a given time series. Assessments of random uncertainty are based on the residual deviations associated with an adequate model for a given time series. The approach also incorporates a treatment of the dynamic response of a measurement system as it pertains to random uncertainty in transient measurements.

The Prediction of Cotton Yarn Irregularity Based on the ‘AFIS’ Measurement

This paper presents a neural network model for predicting the yarn irregularity, based on inputs of fiber property measurements with the AFIS instrument. By using a back-propagation neural network algorithm, alternative models were fitted and compared. The resulting predictions of yarn irregularity are superior to these obtained by using conventional multiple-linear regression techniques.

Distribution feeder loss analysis by using an artificial neural network

This paper proposes an artificial neural network (ANN) based feeder loss analysis for distribution system analysis. The functional-link network model is examined to form the artificial neural network architecture to derive various loss calculation models for distribution feeders with different configurations. The ANN is a feedforward network that uses a standard back-propagation algorithm to adjust the weights on the connection path between any two processing elements. The typical daily load curve of the study feeder for each season is derived to field test data. A three-phase load flow program is then executed to create the ANN training sets to solve the exact feeder loss. A sensitivity analysis is performed to determine the key factors of feeder loss, which are feeder loading and power factor, primary and secondary conductor length, and transformer capacity. The above key factors form the variables of the ANN input layer. By applying the artificial neural network with pattern recognition capability, this study has developed the seasonal loss calculation models for both an overhead and an underground distribution feeder. Two practical feeders in the Taiwan Power Company (Taipower) distribution system have been selected for computer simulation to demonstrate the effectiveness and accuracy of the proposed ANN loss models. By comparing the loss models derived by the conventional regression technique, it is found that the proposed loss models can estimate feeder loss in a very effective manner and provide a better tool for distribution engineers to enhance system operation efficiency.

Small-cell carcinoma of the lung: derivation of a prognostic staging system.

Retrospective data on 22 pretreatment attributes were evaluated in 614 patients with small-cell carcinoma of the lung (SCCL). The series included 284 patients with limited disease (LD) and 328 patients with extensive disease (ED) managed between 1974 and 1986. Prognostic factors were evaluated by univariate analysis and by the Cox multivariate regression model. Recursive partition and amalgamation algorithm (RECPAM), two clustering methods well suited for obtaining strata and adapted for censoring survival data, were developed and used in the formulation of a new prognostic staging system. In univariate analysis, prognosis was significantly influenced by extent of disease (DE), the number of metastatic sites, and the detection of mediastinal spread in LD. Poor performance status (PS), male sex, and advanced age were negatively correlated with survival, as were increased serum levels of alkaline phosphates (AP), lactate dehydrogenase (LDH), carcinoembryonic antigen (CEA), total WBC count (WBCC), and low platelet count and low serum sodium. The Cox model identified plasma LDH and mediastinal spread as the only significant factors in LD; the influence of PS, number of metastatic sites, bone metastasis, brain metastasis, and platelet count were identified as significant in ED. The RECPAM model identified four distinct risk groups defined in a classification tree by the following eight attributes: DE, PS, serum AP, serum LDH, mediastinal spread, sex, WBCC, and liver metastasis. The four groups were distinguished by median survival times of 59, 49, 35, and 24 weeks, respectively (P = .0001). Interactions among prognostic factors are emphasized in the RECPAM classification model as evidenced by reassignment of patients across conventional staging barriers into alternate prognostic groups. The advantages of using RECPAM over the more conventional Cox regression techniques for a new staging system are discussed.

An Updated Comparison of Drug Dosing Methods

The resurgence of the use of and interest in vancomycin, in conjunction with the high degree of interpatient variability in its pharmacokinetic profile, has prompted the development of many and varied dosing methods. Several dosing nomograms have been proposed and evaluated, methods which are useful for initial dosing but do not allow for individualisation of dosage. Given these constraints, several investigators have attempted to apply conventional least-squares regression techniques and, more recently, Bayesian methodologies using either 1- or 2-compartment pharmacokinetic models. Comparative information evaluating algorithmic methods demonstrates that those of Moellering and Lake offer the least biased and most precise predictions of vancomycin dosage. Patient individualisation using conventional least-squares methodology offers some improvement over nomogram-based methods, both in predictive performance and in dosage adjustment once serum concentration data are available. Overall, the latest data indicate that regimens which incorporate Bayesian principles tend to give better results than nomogram-based or conventional least-squares dosing methods for this drug. Despite the advances in methods for dosing vancomycin, several questions remain to be answered. A lack of convincing evidence of a correlation between serum concentrations and therapeutic outcome has prompted debate over the need for serum concentration monitoring and, if it is needed, over which patient population would most benefit. Secondly, little comparative information is currently available as to the dosing of vancomycin in paediatric and neonatal patient populations. Several nomograms for initial dosing have been proposed, but only 2 have been subject to subsequent testing. Finally, information regarding cost-effectiveness and the quality of patient outcome is lacking from the current literature.

Factors influencing the caries experience of a group of children at the ages of 11–12 and 15–16 years: results from an ongoing epidemiological survey

An analysis of factors influencing the caries experience of adolescents in South Wales is presented. Approximately 1000 children were assessed for caries status and oral cleanliness in 1980 when aged 11-12 years and again in 1984 when aged 15-16 years. In addition, on both occasions, the children completed detailed questionnaires on dental health-related topics. When aged 11-12 years, the observed mean DMFT, DMFS and DFS scores of the children were 4.0, 6.7 and 5.5 respectively. The corresponding scores at age 15-16 years were 6.5, 11.8 and 10.2. A preliminary analysis using conventional multiple regression techniques revealed that a number of factors had a significant influence on the caries experience of the children. The significance of the factors depended on the sex of the population subgroup, the age of the children and the particular caries index studied. However, at both ages the factors of most significance were the number of erupted teeth, total mean plaque score and the reported amount of money spent on sweets per week. A further evaluation using analysis of covariance with the number of erupted teeth and surfaces as the covariates confirmed the significant influence of the total mean plaque score and amount of money spent on sweets. In addition, both analyses indicated that toothbrushing frequency and social class had a significant influence on the caries experience of boys.

Modifying nurses' clinical decisions about pain

Shortened versions of various depression scales have typically been based on conventional techniques such as factor analysis, without simultaneously considering the removal of culturally biased items that show differential item functioning (DIF). The authors recently showed that eight items in the 15-item Geriatric Depression Scale (GDS-15) evidenced DIF, associated with gender, age, ethnicity, and chronic illnesses. By selecting out these items, the authors derived a shortened 7-item GDS and compared it with derivations based on conventional factor analysis and logistic regression techniques and validated its test performance in Asian elderly.The GDS-15 and Structured Clinical Interview for Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, diagnosis of major depressive disorder (MDD) was independently administered by interviewers to 4,253 noninstitutionalized community-living elderly subjects aged 60 years and older. Concurrent methods of data reduction and item selection included factor analyses and logistic regression modeling. The GDS-7 was compared with GDS-15 test performance using Structured Clinical Interview for Diagnostic and Statistical Manual of Mental Disorders diagnosis of MDD as gold standard criterion and receiver operating characteristics analysis.After removal of eight DIF items, the derivation was consistent with and complements the results from factor analysis and logistic regression analyses. The short GDS-7 fit a unidimensional model with high explained total variance and demonstrated high sensitivity (0.93) and specificity (0.91) at a cutoff of one-half that was comparable with GDS-15.By removing items with culturally based response bias, the GDS-7 showed excellent scaling and test performance for screening MDD in an Asian elderly population.