Multiple Linear Regression Model For Prediction Research Articles

Gros Michel banana soluble solids content evaluation and maturity classification using a developed portable 6 channel NIR device measurement

In this paper, a portable 6-digital-channel NIR device was designed and implemented for predicting the quality of Gros Michel bananas and to classify them by maturity stage. Our device can be compared to the standard Brix refractometer in the % Brix form. Experimental results indicate that a predictive multiple linear regression model can efficiently determine the banana soluble solids content, with a standard error of estimate of 0.17% and a precision of 0.38% at the yellow stage. In addition, a novel classification of banana’s ripening is also proposed, and the results demonstrate that bananas can automatically and efficiently be classified by maturity level (i.e., to seven ripeness levels) with nearly 100% accuracy. Our approach demonstrated here can also potentially be applied more widely to fruit quality testing with a fundamental framework for future researches.

Acoustics in urban parks: Does the structure of narrow urban parks matter in designing a calmer urban landscape?

Urban parks can function as a proper sink of noise pollution. However, lack of universally-agreed upon methodologies and differing urban conditions have fueled controversy surrounding the effectuation of this urban park function around the world. Hence, to address this controversy in narrow urban parks (with a mean width of ∼109 m) in Isfahan City, Central Iran, noise levels (Lq30) were measured along two longitudinal transects placed along the interior northern river-and adjacent to the southern edge of the parks bordered by a heavily-conjected road. We used statistical tests and models to determine the association of noise levels measured along the northern transect with the distance to, and the intensity of noise emitted from, the road, vegetation biomass, and vegetation height within two 50 and 100 m buffer rings drawn around northern sites and the richness of bird species. The average Lq30 values differed significantly between the southern (∼73.21 dB) and northern (∼66.43 dB) transects and correlated negatively with species richness (r(98) = −0.324, p < 0.01). Three variables including mean NDVIwithin the 100 m buffer ring, distance from the road and mean Lq30 values of the nearest three southern sites were included to build the best predictive multiple-linear regression model through the step-wise procedure with r2 of 0.52. These findings suggest that further attempts aiming to alleviate the parks’ interior noise level should be attentive to distance to road, traffic at the nearest road part, and the interior vegetation characteristics.

The influence of tree traits on urban ground surface shade cooling

The physical nature of the built urban environment gives rise to urban heat islands (UHI), making many cities frequently thermally uncomfortable in the summer, with potentially serious effects on human health. When climate change effects of higher summer temperatures and prolonged heatwaves are factored in, it is clear that adaptive measures are needed to ensure the liveability of cities. The shade provided by planting trees is one such adaptation measure. This study, in Bolzano, Italy, used a thermal camera to record the surface temperatures of three common urban surfaces – asphalt, porphyry, and grass – in the shade of 332 single tree crowns, of 85 different species, during the peak temperature period of summer days. By comparing with the temperature of adjacent unshaded ground, estimates of the degree of surface cooling were made. Measurements at three locations within the shadow revealed higher cooling in the centre and at the western edge. The cooling was related to a multitude of tree traits, of which Leaf Area Index estimate (LAIcept) and crown width were the most important. Median average cooling of 16.4, 12.9 and 8.5 °C was seen in the western edge of the tree shade for asphalt, porphyry and grass, respectively. Maximum temperatures were reduced by roughly 19 °C for all surface types. Coniferous trees were capable of providing high cooling, however, crown dimensions may limit the receiving surface area. Descriptive and predictive multiple linear regression models were able to predict cooling with some success from several of the predictor variables (LAIcept and gap fraction). Strategic planting of single trees in cities can have significant impacts on the absorption of solar radiation by ground surface materials thus reducing the heat storage that contributes to UHIs.

Canopy hyperspectral characteristics and yield estimation of winter wheat (Triticum aestivum) under low temperature injury

To evaluate the effect of low-temperature stress in winter wheat during the early growth stages, the response regularity of the canopy spectral reflectance was evaluated. Besides, winter wheat yield during the maturation stage and the relationship between yield and canopy spectral reflectance were also analyzed. Two multivariate methods, namely, the successive projections algorithm (SPA) and multiple linear regression (MLR), were combined to explore the relationship between the spectral reflectance and yield. Our results showed that the green peak and red valley in visible wavelengths altered obviously and the red edge gradually moved towards blue wavebands. The canopy spectral reflectance in the near-infrared wavebands increased with an increase in low-temperature stress intensity. Moreover, the reflectance proved that the red edge region under low-temperature stress is related to winter wheat yield, and approximately 38% of extracted wavebands were concentrated in the red edge region (680–780 nm). Compared with the predictive MLR models, the model calibrated during the flowering period of winter wheat (25 days post low-temperature treatment) had better performance in predicting crop yield. Whole-spectrum predictive models based on the principle component regression (PCR) method and Normalized Difference Vegetation Index (NDVI) models based on MLR were also established. Moreover, the performance of three kinds of calibration methods and the validation result of the field test were compared to select the optimal monitoring stage and technique to estimate the yield in the early growth stage of winter wheat under low-temperature stress. This study could provide a theoretical basis and practical reference for hyperspectral assessment of yield in winter wheat during low-temperature stress.

Dissolved organic carbon in streams within a subarctic catchment analysed using a GIS/remote sensing approach.

Climate change projections show that temperature and precipitation increases can alter the exchange of greenhouse gases between the atmosphere and high latitude landscapes, including their freshwaters. Dissolved organic carbon (DOC) plays an important role in greenhouse gas emissions, but the impact of catchment productivity on DOC release to subarctic waters remains poorly known, especially at regional scales. We test the hypothesis that increased terrestrial productivity, as indicated by the normalized difference vegetation index (NDVI), generates higher stream DOC concentrations in the Stordalen catchment in subarctic Sweden. Furthermore, we aimed to determine the degree to which other generic catchment properties (elevation, slope) explain DOC concentration, and whether or not land cover variables representing the local vegetation type (e.g., mire, forest) need to be included to obtain adequate predictive models for DOC delivered into rivers. We show that the land cover type, especially the proportion of mire, played a dominant role in the catchment’s release of DOC, while NDVI, slope, and elevation were supporting predictor variables. The NDVI as a single predictor showed weak and inconsistent relationships to DOC concentrations in recipient waters, yet NDVI was a significant positive regulator of DOC in multiple regression models that included land cover variables. Our study illustrates that vegetation type exerts primary control in DOC regulation in Stordalen, while productivity (NDVI) is of secondary importance. Thus, predictive multiple linear regression models for DOC can be utilized combining these different types of explanatory variables.

A hybrid artificial neural network and particle swarm optimization for prediction of removal of hazardous dye brilliant green from aqueous solution using zinc sulfide nanoparticle loaded on activated carbon

In the present study, zinc sulfide nanoparticle loaded on activated carbon (ZnS-NP-AC) simply was synthesized in the presence of ultrasound and characterized using different techniques such as SEM and BET analysis. Then, this material was used for brilliant green (BG) removal. To dependency of BG removal percentage toward various parameters including pH, adsorbent dosage, initial dye concentration and contact time were examined and optimized. The mechanism and rate of adsorption was ascertained by analyzing experimental data at various time to conventional kinetic models such as pseudo-first-order and second order, Elovich and intra-particle diffusion models. Comparison according to general criterion such as relative error in adsorption capacity and correlation coefficient confirm the usability of pseudo-second-order kinetic model for explanation of data. The Langmuir models is efficiently can explained the behavior of adsorption system to give full information about interaction of BG with ZnS-NP-AC. A multiple linear regression (MLR) and a hybrid of artificial neural network and partial swarm optimization (ANN–PSO) model were used for prediction of brilliant green adsorption onto ZnS-NP-AC. Comparison of the results obtained using offered models confirm higher ability of ANN model compare to the MLR model for prediction of BG adsorption onto ZnS-NP-AC. Using the optimal ANN–PSO model the coefficient of determination (R2) were 0.9610 and 0.9506; mean squared error (MSE) values were 0.0020 and 0.0022 for the training and testing data set, respectively.

Correlations of limiting oxygen index with structural polyphosphoester features by QSPR approaches

Fire retardant materials diminish the hazard to life from fire. Polyphosphonates and polyphosphates display good flame retardancy and attractive plasticizing properties, being an important class of organophosphorus based polymer additives. Properties of previously synthesized polyphosphoesters are presented here, being simulated as monomers. In this quantitative structure–property relationship work, the flammability was expressed by limiting oxygen index (LOI) values, which were determined experimentally. Two types of chiral structures were found by molecular mechanics calculations using the MMFF94s force field for half of the monomers, consequently, two datasets were built. Structural parameters were calculated for the minimum energy structures and were related to the LOI values by multiple linear regression (MLR), artificial neural networks (ANNs), and support vector machines (SVMs). MLR calculations were combined with a genetic algorithm for variable selection. Stable and predictive MLR models in terms of 2D autocorrelation parameters weighed by atomic polarizabilities and of 3D-Morse descriptors were obtained. Somewhat inferior fits were found in the nonlinear modeling by ANNs and SVMs with the same set of descriptors. Monomers including R chiral structures gave more stable and predictive models compared to the S isomers in all approaches. Monomer geometry influences the flame retardancy, being favorable for R isomers.

Artificial neural network and particle swarm optimization for removal of methyl orange by gold nanoparticles loaded on activated carbon and Tamarisk

The influence of variables, namely initial dye concentration, adsorbent dosage (g), stirrer speed (rpm) and contact time (min) on the removal of methyl orange (MO) by gold nanoparticles loaded on activated carbon (Au-NP-AC) and Tamarisk were investigated using multiple linear regression (MLR) and artificial neural network (ANN) and the variables were optimized by partial swarm optimization (PSO). Comparison of the results achieved using proposed models, showed the ANN model was better than the MLR model for prediction of methyl orange removal using Au-NP-AC and Tamarisk. Using the optimal ANN model the coefficient of determination (R2) for the test data set were 0.958 and 0.989; mean squared error (MSE) values were 0.00082 and 0.0006 for Au-NP-AC and Tamarisk adsorbent, respectively. In this study a novel and green approach were reported for the synthesis of gold nanoparticle and activated carbon by Tamarisk. This material was characterized using different techniques such as SEM, TEM, XRD and BET. The usability of Au-NP-AC and activated carbon (AC) Tamarisk for the methyl orange from aqueous solutions was investigated. The effect of variables such as pH, initial dye concentration, adsorbent dosage (g) and contact time (min) on methyl orange removal were studied. Fitting the experimental equilibrium data to various isotherm models such as Langmuir, Freundlich, Tempkin and Dubinin–Radushkevich models show the suitability and applicability of the Langmuir model. Kinetic models such as pseudo-first order, pseudo-second order, Elovich and intraparticle diffusion models indicate that the second-order equation and intraparticle diffusion models control the kinetic of the adsorption process. The small amount of proposed Au-NP-AC and activated carbon (0.015g and 0.75g) is applicable for successful removal of methyl orange (>98%) in short time (20min for Au-NP-AC and 45min for Tamarisk-AC) with high adsorption capacity 161mgg−1 for Au-NP-AC and 3.84mgg−1 for Tamarisk-AC.

Artificial neural network and Bees Algorithm for removal of Eosin B using Cobalt Oxide Nanoparticle‐activated carbon: Isotherm and Kinetics study

The objective of this work is the study of adsorption of Eosin B by cobalt oxide nanoparticle loaded on activated carbon (Co2O3‐NP‐AC). This new material with high efficiency in a routine manner was synthesized in our laboratory, and its surface properties such as surface area, pore volume, and functional groups were characterized with different techniques such X‐ray diffraction, Brunauer, Emmett, and Teller, and scanning electron microscopy analysis. The effect of solution pH, adsorbent dosage (0.005–0.02 g), contact time (0.5–30 min), and initial concentration of dye (30–80 mg L−1) on the adsorption process was investigated. Thus, Langmuir, Freundlich, Tempkin, and D–R isothermal models are applied for fitting the experimental data, and the data well presented by Langmuir model with a maximum adsorption capacity of 588.2 mg g−1 at 25°C. Kinetic studies at various adsorbent dosage and initial Eosin B concentration show that maximum Eosin B removal was achieved within 18 min of the start of every experiment at most conditions. The combination of pseudo‐second‐order rate equation and intraparticle diffusion model (with removal more than 99%) is usable to explain the experimental data of adsorption process at all conditions. The influences of parameters including initial dye concentration, adsorbent dosage (g), and contact time on Eosin B adsorption onto cobalt oxide nanoparticles loaded on AC were investigated by multiple linear regression (MLR) and artificial neural network (ANN), and the influences of variables were optimized using Bees Algorithm. Comparison of the results obtained using introduced models showed the ANN model is better than the MLR model for prediction of Eosin B removal using cobalt oxide nanoparticles loaded on AC. Using the optimal ANN model, the coefficients of determination (R2) were 0.9965 and 0.9936; mean squared error values were 0.00015 and 0.00029 for training and testing data, respectively. © 2014 American Institute of Chemical Engineers Environ Prog, 34: 155–168, 2015

Transition from a predictive multiple linear regression model to an explanatory simple nonlinear regression model with higher level of prediction: A systems dynamics approach

One of the main assumptions of the linear regression analysis is the existence of a causal relationship between the variables analyzed, which the regression analysis does not demonstrate. This paper demonstrates the causality between the variables analyzed through the construction and analysis of the feedback from the variables under study, expressed in a causal diagram and validated through dynamic simulation. The major contribution of this research is the proposal of the use of the system dynamics approach to develop a method of transition from a multiple regression predictive model to a simpler nonlinear regression explanatory model, which increases the level of prediction of the model. The mean square error (MSE) is taken as a criterion for prediction. The validation in the transition model was performed with three linear regression models obtained experimentally in a textile company, showing a method for increasing the reliability of prediction models.

Obesity-independent Association of Human Adenovirus Ad37 Seropositivity With Nonalcoholic Fatty Liver Disease

Adenoviruses Ad36 and Ad37 increase adiposity in animals and are associated with obesity in humans; effects on the liver have been reported. The association of Adenovirus Ad36 seropositivity (Ad36+) with obesity but not with the severity of nonalcoholic fatty liver disease (NAFLD) has been previously shown. We investigate whether nondiabetic Ad37+ patients show a different prevalence of NAFLD and ultrasound Bright Liver score. A total of 268 adult nondiabetic patients (146 men, 122 women) were included after lifestyle counseling including a personalized Mediterranean diet, increase in physical activity, and smoking withdrawal. After an Ad37+/Ad36+ assay, overweight obesity, insulin resistance, C-reactive protein, and bright liver prevalence and severity were compared according to Ad37+. Sixty-five of 268 patients were Ad37+ and 82/268 patients were both Ad37 seronegative (Ad37-) and Ad36-. The prevalence of obesity, defined as body mass index≥30, was not significantly different in Ad37+ (11/65; 16.9%) vs. Ad37- (15/82; 18.2%) patients; Bright Liver was present in 22/65 (33.8%) Ad37+ patients vs. 13/82 (15.8%) Ad37- patients (P<0.019). By odds ratio (OR), a consistent risk for NAFLD was associated with Ad37+, greater insulin resistance, and C-reactive protein. By a predictive multiple linear regression model, 40.0% of variance toward NAFLD and 50.4% toward the severity of Bright Liver score was explained significantly and independently by Ad37+ and by body mass index. Ad37+ status in nondiabetic patients on an appropriate diet is significantly associated with NAFLD; because fatty liver improves even without weight loss by a "healthy" diet, and not only by lower food caloric intake, Ad37+ may be an adjunctive hallmark of an unfavorable clinical-metabolic profile, if not a causative factor of NAFLD.

Correction methods for statistical models in tropospheric ozone forecasting

Abstract This study proposes two methods to enhance the performance of statistical models for prediction tropospheric ozone concentrations. The first method corrects the statistical model based on the average daily profile of the model errors in training set. The second method estimates the model error by making the analogy with three basic modes of feedback control: proportional, integral and derivative. These correction methods were tested with multiple linear regression (MLR) and artificial neural networks (ANN) for prediction of hourly average tropospheric ozone (O3) concentrations. The inputs of the models were the hourly average concentrations of sulphur dioxide (SO2), carbon monoxide (CO), nitrogen oxide (NO), nitrogen dioxide (NO2) and O3, and some meteorological variables (temperature – T; relative humidity – RH; and wind speed – WS) measured 24 h before. The analysed period was from May to June 2003 divided in training and test periods. ANN presented slightly better performance than MLR model for prediction of O3 concentrations. Both models presented improvements with the proposed correction methods. The first method achieved the highest improvements with ANN model. However, the second method was the one that obtained the best predictions of hourly average O3 concentrations with the correction of MLR model.

Are Mechanistic and Statistical QSAR Approaches Really Different? MLR Studies on 158 Cycloalkyl-Pyranones.

Two parallel approaches for quantitative structure-activity relationships (QSAR) are predominant in literature, one guided by mechanistic methods (including read-across) and another by the use of statistical methods. To bridge the gap between these two approaches and to verify their main differences, a comparative study of mechanistically relevant and statistically relevant QSAR models, developed on a case study of 158 cycloalkyl-pyranones, biologically active on inhibition (Ki ) of HIV protease, was performed. Firstly, Multiple Linear Regression (MLR) based models were developed starting from a limited amount of molecular descriptors which were widely proven to have mechanistic interpretation. Then robust and predictive MLR models were developed on the same set using two different statistical approaches unbiased of input descriptors. Development of models based on Statistical I method was guided by stepwise addition of descriptors while Genetic Algorithm based selection of descriptors was used for the Statistical II. Internal validation, the standard error of the estimate, and Fisher's significance test were performed for both the statistical models. In addition, external validation was performed for Statistical II model, and Applicability Domain was verified as normally practiced in this approach. The relationships between the activity and the important descriptors selected in all the models were analyzed and compared. It is concluded that, despite the different type and number of input descriptors, and the applied descriptor selection tools or the algorithms used for developing the final model, the mechanistical and statistical approach are comparable to each other in terms of quality and also for mechanistic interpretability of modelling descriptors. Agreement can be observed between these two approaches and the better result could be a consensus prediction from both the models.

Analysis of auditory measures in normal hearing young male adult cigarette smokers using multiple variable selection methods with predictive validation assessments.

Studies have shown that cigarette smoking is a risk factor for hearing loss; however, no information is available on auditory preclinical indicators in young chronic cigarette smokers. Cigarette smoking involves exposure to many harmful chemicals including carbon monoxide (CO). In this study, the CO level in 16 young normal hearing male chronic smokers was measured with a CO monitor, and was used as the outcome measure. Subjects were administered a battery of audiological tests that included behavioral and electrophysiologic measures. The goal was to investigate which auditory test measures can be used as potential predictors of the outcome measure. Using ordinary least squares estimation procedures with best-subsets selection and bootstrapped stepwise variable selection procedures, an optimal predictive multiple linear regression model was selected. Results of this approach indicated that auditory brainstem response peak V amplitudes and distortion product otoacoustic emissions had the highest predictive value and accounted for most of the variability.

Direct and Indirect Quantum Mechanical QSPR Hildebrand Solubility Parameter Models

AbstractThe objectives of this study were to develop Quantitative Structure–Property Relationship (QSPR) models to predict Hildebrand solubility parameters (δt). The study investigated direct δt prediction by QSPR and indirect δt calculation from QSPR‐predicted enthalpy of vaporization values. Molecular ground‐state minimum energy conformations were identified and optimized using the AM1 semiempirical Quantum Mechanical (QM) method. Predictive multiple linear regression models were derived correlating training set experimental solubility parameters or enthalpies of vaporization with descriptors obtained from the QM calculations. A four‐descriptor Hildebrand total solubility parameter regression equation was developed where R2=0.97, R$\rm{ {_{adj}^{2}}}$=0.97, F=461.5, s2=0.53, and Root Mean Square Error (RMSE)=0.69. A four‐descriptor QSPR model to predict enthalpies of vaporization (ΔHvap−kJ/mol) was constructed where R2=0.96, R$\rm{ {_{adj}^{2}}}$=0.96, F=230.3, s2=4.75, and RMSE=2.04. δt calculations from ΔHvap correlated well with literature δt , where R2=0.96. Both of these models are statistically significant as well as predictive. The direct and indirect δt models developed here compare favorably with liquid solubility parameter estimation accuracies (10%) reported by Hoftyzer–Van Krevelen and Hoy. These methods are useful for determining the miscibility of small organic molecules, an important consideration when formulating polymer blends.

Long-Range Prediction of the Shipping Season in Hudson Bay: A Statistical Approach

Abstract Despite recent reductions in Arctic sea ice extent and the associated increase in both the recreational and commercial use of ice-infested waters, long-range prediction of operationally relevant sea ice parameters is an area of seasonal forecasting that has received little attention. Statistical methods that isolate and exploit empirical relationships between antecedent low-frequency climate variability and specific variables of interest are often used to solve seasonal forecasting problems. In this study, simple multiple linear regression (MLR) techniques are used to improve the skill of the seasonal (3-month lead) forecast of the breakup and clearing of sea ice along the shipping route through Hudson Bay that is issued each March by the Canadian Ice Service of Environment Canada. Using sea ice and climate data from 1972 to 2002, predictive MLR models are developed for the spring opening date of the shipping route and the latest expected opening date. A success rate of 77% over the 1972–2002 period for the opening date, from an MLR model that explains 76% of the variability in the original time series with a mean absolute error (MAE) of 0.38, is a marked improvement over the 48% success rate of the current analog methodology. The success rate of the model for the latest expected date is 87%; the modeled time series adequately represented interannual variability in the observed time series (r = 0.71) with a low MAE (MAE = 0.51). Results from a series of model diagnostics that include Monte Carlo simulations, cross validation, and analysis of residuals, suggest the final models are statistically valid and are not influenced by artificial skill. The main source of predictive skill in the model is winter low-frequency variability in North Atlantic sea surface temperatures and 500-mb geopotential heights; physical processes that may explain this link are presented. It is concluded that simple multiple linear regression techniques can be applied to generate use-specific seasonal forecasts of sea ice conditions and that the empirical knowledge gained in the model development may help elucidate or identify physical processes in the climate system.