Principal Component Regression Model Research Articles

Artificial intelligence-based prediction of lycopene content in raw tomatoes using physicochemical attributes.

Lycopene consumption reduces risk and incidence of cancer and cardiovascular diseases. Tomatoes are a rich source of phytochemical compounds including lycopene as a major constituent. Lycopene estimation using high-performance liquid chromatography is time-consuming and expensive. To develop artificial intelligence models for prediction of lycopene in raw tomatoes using 14 different physicochemical parameters including salinity, total dissolved solids (TDS), electrical conductivity (EC), firmness, pH, total soluble solids (TSS), titratable acidity (TA), colour values on Hunter scale (L, a, b), total phenolic content (TPC), total flavonoid content (TFC) and antioxidant activity (AOA). The post-harvest data acquisition was collected through investigation for more than 100 raw tomatoes stored for 15 days. Linear multivariate regression (LMVR), principal component regression (PCR) and partial least squares regression (PLSR) models were developed by splitting data set into train and test datasets. The training of models was performed using 10-fold cross validation (CV). Principal component analysis showed strong positive association between lycopene, colour value 'a', TPC, TFC and AOA. The R2 (CV), root mean square error (RMSE) (CV) and RMSE (Test) for best LMVR model was observed to be at 0.70, 8.48 and 9.69 respectively. The PCR model revealed R2 (CV) at 0.59, RMSE (CV) at 8.91 and RMSE (Test) at 10.17 while PLSR model revealed R2 (CV) at 0.60, RMSE (CV) at 9.10 and RMSE (Test) at 10.11. Results of the present study show that epidemiological studies suggest fully ripened tomatoes are most beneficial for consumption to ensure recommended daily intake of lycopene content.

Incorporating field-based research into remote learning: An assessment of soil lead pollution in different land-use types in Los Angeles

A research-based course was developed to investigate the legacy of soil lead (Pb) pollution in Los Angeles, California. During the course, undergraduate and graduate students collected a total of 270 soil samples for analyses of metal (loid) concentrations in different land-use types (residential, park, and school). Residential soils had significantly higher Pb concentrations than other land uses (p < 0.01) exceeding the California recommended safety level for soil Pb (80 mg/kg) at the highest frequency (64% of samples), followed by schools (42%) and parks (6.0%). Soil Pb from all 87 census block groups was correlated with battery recycling plant and railroad proximity as geospatial indicators of childhood Pb exposure risk. Meanwhile, census block groups with higher Pb levels were correlated with higher percentages of the following population: those without health insurance, without college degrees, with a lower median household income and income below the poverty line, and ethnic and racial minorities (r = −0.46 to 0.59, p < 0.05). Principal component regression models significantly improved soil Pb estimation over correlation analysis by incorporating sociodemographic, economic, and geospatial risk factors for Pb exposure (R2 = 0.58, p < 0.05). This work provides new insights into how topsoil Pb prevails in various land-use types and their co-occurring sociodemographic, economic, and geospatial risk factors, indicating the need for multi-scalar assessment across urban land uses.

PC-DOT: Improving genomic prediction ability of principal component regression by DOT product.

Principal component regression (PC regression) is a useful method developed for prediction based on a dimension-reducing strategy. Generally, the principal components (PCs) are added to the regression model one by one based on the eigenvalue (PC-Eigen). Considering that some PCs with large eigenvalues may be poorly associated with the response variable, the PC-Eigen may not be the best framework. Researchers previously tried to add PCs to the model based on their contribution to the sum of squares of the regression (PC-SS) and they found that the performance of PC-SS is generally lower than that of the PC-Eigen. A standard approach for selecting the optimal set of PCs remains a challenge. Here, from the cosine similarity theory, we postulated that we could rank the PCs by dot product, and this framework (we called PC-DOT) could help to preferentially extract PCs that are highly correlated with the response variable and meanwhile have a large eigenvalue. Based on one simulated and three real genomic datasets (a total of 15 traits), we tested the prediction ability of different frameworks. In general, the PC-DOT method showed a better performance than both PC-Eigen and PC-SS. To facilitate the application of PC, we attached a series of R codes for different frameworks (https://github.com/SUNHAO-JLU/Genome_Prediction-PC_DOT). In addition, the HAT matrix was used to reduce the compute complex in reference data during the cross-validation process. Our work may help researchers to better understand and carry out the PC regression model.

A novel methodology to integrate outcomes regarding perioperative pain experience into a composite score: Prediction model development and validation

An integrated score that globally assesses perioperative pain experience and rationally weights each component has not yet been developed. A development dataset specific to adult Chinese patients undergoing orthopaedic surgery was obtained from PAIN OUT (1985 qualified patients of 2244). A more recent validation dataset obeying the same conditions was obtained from the Chinese Anaesthesia Shared-database Platform (1004 qualified patients of 1032). Outcomes were assessed using the International Pain Outcomes Questionnaire (IPO-Q), which comprises key patient-level outcomes of perioperative pain management, including pain experience and perceptions of care. Using principal component analysis and regression models, a composite score (CS) was inferred to integrate pain experience. The discrimination of the CS for dissatisfaction and desire for more pain treatment was compared with that of the worst pain score. A CS was developed from the 12 items of the IPO-Q regarding pain experience. The weight for calculating the CS was worst pain 11, least pain 17, time spent in severe pain 11, interference with activity in bed 9, interference with breathing deeply or coughing 10, interference with sleep 9, anxiety 12, helplessness 12, nausea 0, drowsiness 2, itch 5 and dizziness 2. In external validation, the CS indicated superior discrimination to the worst pain in predicting dissatisfaction (p&lt; 0.001) and desire for more pain treatment (p&lt; 0.001). This study introduced a methodology to integrate outcomes regarding perioperative pain experience into a CS, which was based on the weight of each item. This novel methodology sheds additional light on the riveting issue of carefully integrating several measures into a composite endpoint, which may be useful for quality improvement purposes when addressing the impact of a change in clinical practice.

Changes and relationships of viscoelastic and physical properties of Chinese yam during a novel multiphase microwave drying process

Changes in viscoelastic properties and physical qualities including rehydration ratio (RR), volume ratio (Vr), porosity and texture of Chinese yams during multiphase microwave drying (MMD) were studied and the intrinsic relationship between them was analyzed. Results showed that RR, Vr and porosity rose and then fell with drying time during MMD, with the endpoint of sublimation drying being the turning point. Scheme III with the highest microwave power (0.2 W/g) in sublimation drying resulted in the lowest RR, Vr and porosity but the highest hardness. The Maxwell model was well fitted to describe the viscoelastic behaviors of yams (R2 > 0.99). Enhanced stress relaxation behavior during sublimation drying was observed. In contrast, a decrease in stress decay but an increase in residual elasticity were found when MMD switched from sublimation drying to evaporation drying. Pearson's correlation analysis showed that relaxation modulus was significantly negatively correlated with RR (r = −0.981, p < 0.01) and porosity (r = −0.961), while positively correlated with hardness (r = 0.94, p < 0.05) and the brittleness (r = 0.96, p < 0.01). Principal component regression models between the viscoelastic indexes and physical features with high R2 (>0.9) values could explain their relationships well. These findings have implications for the production of high-quality products through controlling drying processes.

Advanced Spectrophotometric Resolution Techniques for a Spectrally Overlapping Spasmolytic Binary Mixture along with Dichloroaniline as a Toxic Impurity: Application to Content Uniformity Testing.

Irritable bowel syndrome (IBS) is a common disorder leading to undesirable pain. Phloroglucinol (PHG) and trimethylphloroglucinol (TMG) are co-formulated as spasmolytic medication that is considered to be effective in reducing smooth muscle spasm. 3,5-Dichloroaniline (DCL) is a specified PHG pharmacopoeial impurity which needs to be monitored to avoid its toxic effects. Different smart approaches are presented to provide simple, reliable, and economic spectrophotometric methods able to resolve the severe overlap in the spectra of PHG and TMG in their pure and pharmaceutical forms, in addition to their estimation in the presence of DCL as a toxic impurity of PHG without any need for initial separation. The presented work includes univariate methods, derivative ratio (DR), ratio difference (RD), mean centering (MCR) and deconvulated Fourier method (DF), which were able to determine PHG and TMG simultaneously in their binary mixture. Firstly, DCL was estimated in the zero order, where the two drugs have zero absorption at 247.0 nm, and then its contribution was eliminated by applying ratio subtraction method. Multivariate chemometric partial least squares (PLS) and principal component regression (PCR) models were also applied to determine PHG and TMG simultaneously in the presence of the impurity, DCL. Univariate methods were applied in the range 5.0-30.0, 2.5-25.0, and 1.0-12.0 µg/mL for PHG, TMG, and DCL, respectively. The proposed chemometric models were used in the range 6.0-14.0, 5.0-25.0 and 2.0-10.0 µg/mL for PHG, TMG, and DCL, respectively. These analytical approaches succeeded in estimating the cited drugs in their pharmaceutical formulation and assessing content uniformity of dosage units. The methods were statistically compared with a reported HPLC method, and the results revealed no significance statistical difference. This work provides for the first time successful univariate and multivariate PLS and PCR methods to assess PHG and TMG in the presence of DCL as a toxic impurity along with content uniformity testing of dosage units. Comparative univariate and multivariate spectrophotometric analytical approaches are presented, for the first time, for estimation of spasmolytic formulation of PHG and TMG in the presence of DCL as a PHG toxic impurity. Successful application to content uniformity testing of Stopspasm® dosage form is demonstrated. A statistical study, including t-tests and one-way analysis of variance (ANOVA), was conducted.

Development and Validation of a Vibration-Based Virtual Sensor for Real-Time Monitoring NOx Emissions of a Diesel Engine

With a strong legal basis and regulatory authority, cost-effective transient emission sensors that reflect real-driving emissions are key factors for accomplishing environmental requirements. It is difficult for the existing NOx emission monitoring techniques to achieve a balance between accuracy and timeliness. Fundamentally, in-cylinder combustion is the thermodynamic cause of NOx emissions and the main excitation source for engine vibration and noise emissions. A novel vibration-based virtual NOx sensor is developed based on these critical relationships for real-time NOx monitoring. First, the correlation between vibration and NOx emission was characterized in-depth. Then, a technique of constructing two-dimensional filters for vibration signals is proposed to extract combustion-related information. A principal component regression (PCR) model for NOx prediction was established based on the reconstructed in-cylinder pressure. Finally, the virtual NOx sensor is tested and validated on a single-cylinder diesel engine bench. The virtual NOx sensor is proven to meet the accuracy requirement of vehicle emission monitoring for both steady-state and transient conditions and has a better frequency response compared to the emission measurement system.

Eye Tracking Metrics Differences among Uninjured Adolescents and Those with Acute or Persistent Post-Concussion Symptoms.

Eye tracking assessments that include pupil metrics can supplement current clinical assessments of vision and autonomic dysfunction in concussed adolescents. This study aimed to explore the utility of a 220-second eye tracking assessment in distinguishing eye position, saccadic movement, and pupillary dynamics among uninjured adolescents, those with acute post-concussion symptoms (≤28 days since concussion), or those with persistent post-concussion symptoms (>28 days since concussion). Two hundred fifty-six eye tracking metrics across a prospective observational cohort of 180 uninjured adolescents recruited from a private suburban high school and 224 concussed adolescents, with acute or persistent symptoms, recruited from a tertiary care subspecialty concussion care program, 13 to 17 years old, from August 2017 to June 2021 were compared. Kruskal-Wallis tests were used, and Bonferroni corrections were applied to account for multiple comparisons and constructed receiver operating characteristic curves. Principal components analysis and regression models were applied to determine whether eye tracking metrics can augment clinical and demographic information in differentiating uninjured controls from concussed adolescents. Two metrics of eye position were worse in those with concussion than uninjured adolescents, and only one metric was significantly different between acute cases and persistent cases. Concussed adolescents had larger left and right mean, median, minimum, and maximum pupil size than uninjured controls. Concussed adolescents had greater differences in mean, median, and variance of left and right pupil size. Twelve metrics distinguished female concussed participants from uninjured; only four were associated with concussion status in males. A logistic regression model including clinical and demographics data and transformed eye tracking metrics performed better in predicting concussion status than clinical and demographics data alone. Objective eye tracking technology is capable of quickly identifying vision and pupillary disturbances after concussion, augmenting traditional clinical concussion assessments. These metrics may add to existing clinical practice for monitoring recovery in a heterogeneous adolescent concussion population.

Critical assessment of smart calculation-based spectroscopy versus chemometric-assisted methods: Application to combined antibiotic formulations

This work describes a comparative study of two multivariate chemometric and univariate spectrophotometric methods for the determination of a ternary drug mixture containing oxytetracycline HCl, bromhexine HCl, and lidocaine HCl. All methods show high sensitivity and similar linearity range. Meanwhile, the chemometric method has the advantage of higher accuracy, higher specificity and better regression parameters. The two spectrophotometric methods are constant multiplication coupled with spectrum subtraction and successive ratio subtraction coupled with spectrum subtraction while the chemometric method used partial least square and principal component regression models. In addition, a spiking technique was used to increase the concentration of bromhexine HCl in the dosage form, allowing its determination despite its low contribution. Methods were successfully applied in the dosage form Oxyclear® veterinary injection in pure powder as well as in its pharmaceutical formulation. Statistical comparison showed no significant difference between the developed methods and the reference method.

Computed tomographic evaluation of the acetabulum for age estimation in an Indian population using principal component analysis and regression models

The acetabulum presents as a well-preserved evidence, resistant to taphonomic degradation changes and can thus aid in the age estimation process. A CT-based examination of the acetabulum can further help simplify the process of age estimation by overcoming the time-consuming process of maceration and by doing away with the interference resulting from tissue remnants. The aim of the present study was to evaluate the role of the acetabulum for age estimation in an Indian population through a CT-based examination, using principal component analysis and regression models. CT images of 400 individuals aged 10years and above were evaluated according to the features defined in the San-Millán-Rissech method of age estimation. Five of the seven morphological features defined by San-Millán-Rissech were appreciable on CT scans, and, to enable further statistical analysis, a cumulative score was computed using these five features. A significant correlation of 0.835 and 0.830 for the right and left acetabulum, respectively, was obtained between computed cumulative scores and chronological age of individuals. No significant sex differences were observed in the scoring of different age-related morphological changes. Regression models were generated using individual features and cumulative scores. Regression models derived using the cumulative score yielded inaccuracy values of 9.67years for the right acetabulum and 9.15years for the left acetabulum. Inaccuracy and bias values were computed for each individual feature, as well as for each decade, using mean point ages established within the original study. Amongst the various features, acetabular rim porosity was seen to have the lowest values of inaccuracy (11.50years) and bias (2.32years) and activity on outer edge of acetabular fossa the highest (inaccuracy and bias values of 22.36years and 21.50years, respectively). Taking into consideration this differential contribution towards age estimation, weighted coefficients and mean point ages for different morphological features were determined using principal component analysis. Subsequently, summary age models were generated from the obtained weighted coefficients and mean age values. Summary age models derived in the present study yield lower estimates of inaccuracy of 7.60years for the right acetabulum and 7.82years for the left acetabulum. While regression models derived in the present study allow for age estimation using even a single appreciable feature, summary age models take into account the contribution of each feature and generate more accurate estimates of age. Both statistical computations yield reduced error rates and thus can render greater applicability to the acetabulum in forensic age estimation.

A New Methodology to Produce More Skillful United States Cool-Season Precipitation Forecasts

Abstract The water resources of the western United States have enormous agricultural and municipal demands. At the same time, droughts like the one enveloping the West in the summer of 2021 have disrupted supply of this strained and precious resource. Historically, seasonal forecasts of cool-season (November–March) precipitation from dynamical models such as North American Multi-Model Ensemble (NMME) and the Seasonal Forecasting System 5 (SEAS5) from the European Centre for Medium-Range Weather Forecasts have lacked sufficient skill to aid in Western stakeholders’ and water managers’ decision-making. Here, we propose a new empirical–statistical framework to improve cool-season precipitation forecasts across the contiguous United States (CONUS). This newly developed framework is called the Statistical Climate Ensemble Forecast (SCEF) model. The SCEF framework applies a principal component regression model to predictors and predictands that have undergone dimensionality reduction, where the predictors are large-scale meteorological variables that have been prefiltered in space. The forecasts of the SCEF model captures 12.0% of the total CONUS-wide standardized observed variance over the period 1982/83–2019/20, whereas NMME captures 7.2%. Over the more recent period 2000/01–2019/20, the SCEF, NMME, and SEAS5 models respectively capture 11.8%, 4.0%, and 4.1% of the total CONUS-wide standardized observed variance. An important finding is that much of the improved skill in the SCEF, with respect to models such as NMME and SEAS5, can be attributed to better forecasts across most of the western United States.

Association between maternal psychological factors and offspring executive function: analysis of African-American mother-child dyads.

Executive function (EF) develops throughout childhood and adolescence; however, little is known about whether and how early life factors are associated with EF during these two stages. This secondary analysis examined the associations between maternal psychological characteristics at 2 years after childbirth and offspring EF at 6 and 18 years. Data were from the 18-year New Mothers' Study in Memphis, TN. Women who self-identified as African-American were included (mother-child dyads: N = 414). Maternal psychological characteristics (e.g., depressive symptoms, self-esteem) were assessed using standardized questionnaires; offspring EF at 6 (i.e., working memory, response inhibition) and 18 years (e.g., working memory, sustained attention) were assessed using age-appropriate cognitive tasks. Statistical analyses included principal component analysis (PCA) and regression models. PCA reduced the correlated psychological characteristics to two factors: emotionality (depressive symptoms, emotional instability) and psychological resources (self-esteem, mastery, active coping). After controlling for maternal IQ, maternal emotionality was associated with worse working memory and response inhibition (marginally significant) at 6 years. Maternal psychological resources were marginally associated with better working memory at 6 years. Maternal psychological characteristics may be associated with later EF in offspring. Future studies are needed to replicate these findings and to explore potential mediators. African-American mothers' depressive symptoms and emotional instability at 2 years after childbirth were associated with offspring executive function at 6 and 18 years. African-American mothers' psychological resources at 2 years after childbirth were marginally associated with offspring working memory at 6 years. Maternal IQ attenuated all of the associations observed between maternal psychological status and offspring executive function.

Seasonal forecasting of tropical cyclones over the Bay of Bengal using a hybrid statistical/dynamical model

AbstractThe post‐monsoon (October–November–December) tropical cyclone (TC) over the Bay of Bengal is one of the most devastating natural disasters causing economic and human losses over India and its neighbouring countries. This study discusses a hybrid statistical/dynamical model developed to forecast the post‐monsoon cyclone activities over the Bay of Bengal, where 80% of the TCs of the North Indian Ocean are originated. In the hybrid model, the coupled model CFSv2 predicts the large‐scale climate indices, and the principal component regression (PCR) model is used to relate these indices with the TC frequency. A solid concurrent relation between the cyclonic disturbance frequencies and various large‐scale variables is noted. The dynamical variable, for example, the zonal wind, acts as a precursor variable. We identified three concurrent predictors (ocean heat content over the Bay of Bengal, sea surface temperature (SST) over the Indian Ocean, and SST over the tropical central Pacific regions) and two precursor predictors (low‐level wind at equatorial Indian ocean and strength of upper‐level easterly jet over African coast) influencing the cyclonic disturbance frequencies over the Bay of Bengal. The concurrent predictors are calculated from the CFSv2 hindcast/forecast output and the precursor predictors are calculated from the reanalysis data. The predictors influencing the cyclonic disturbance over the Bay of Bengal are also influencing the cyclonic storms. Hence, the same predictors are used for developing a hybrid model for cyclonic disturbance and storm frequencies. A significant inter‐correlation among different predictors is observed and the PCR model avoids these inter‐correlations and, in this method, PCs are estimated on the predictors to make them orthogonal to each other. The hybrid model achieved a significant skill for seasonal cyclone forecast over the Bay of Bengal. Results suggest the potential for using the hybrid model for the operational seasonal forecasting of post‐monsoon cyclone activity over the Bay of Bengal.

Bayesian inference of the cumulative logistic principal component regression models

Bayesian inference of the cumulative logistic principal component regression models

Predicting engine fuel properties of biodiesel and biodiesel-diesel blends using spectroscopy based approach

Biodiesel can be produced from several feedstocks, resulting in significant composition and fuel properties variations. Predicting fuel properties can significantly reduce engine development time with biodiesel produced from various feedstocks. The applicability of spectroscopy-based support vector regression (SVR) models to predict biodiesel properties are established in the literature. However, similar models for biodiesel-diesel blends are not available. SVR models were developed in the present study based on Fourier Transform Infrared (FTIR) spectra of biodiesel and biodiesel-diesel blends to predict important engine fuel properties. Additionally, SVR models were developed to predict biodiesel blend proportion in biodiesel-diesel blends based on their spectral data. Models were calibrated with 120 samples, comprising 50 biodiesel-diesel blends with 5 to 95% (v/v) biodiesel and 70 biodiesels. The models were validated using 60 samples, including 27 biodiesel-diesel blends and 33 biodiesels. The developed models resulted in a MAPE of 3%, 2.1%, 3%, and 1.19%, respectively, in predicting the blend proportion, viscosity, cetane number, and calorific value. Further, the prediction performance of the SVR models is compared with the conventional principal component regression and partial least square regression models, whose outcome suggest SVR as the most suitable approach.

Application of two statistical approaches (Bayesian Kernel Machine Regression and Principal Component Regression) to assess breast cancer risk in association to exposure to mixtures of brominated flame retardants and per- and polyfluorinated alkylated substances in the E3N cohort

BackgroundBrominated flame retardants (BFR) and per- and polyfluorinated alkylated substances (PFAS) are two groups of substances suspected to act as endocrine disruptors. Such substances could therefore be implicated in the occurrence of breast cancer, nevertheless, previous studies have led to inconstant results. Due to the large correlation between these substances, and the possibly non-linear effects they exert, evaluating their joint impact as mixtures on health remains challenging.This exploratory study aimed to generate hypotheses on the relationship between circulating levels of 7 BFR (6 polybrominated diphenyl ethers and 1 polybrominated biphenyls) and 11 PFAS and the risk of breast cancer in a case–control study nested in the E3N French prospective cohort by performing two methods: Principal Component Regression (PCR) models, and Bayesian Kernel Machine Regression (BKMR) models.Methods194 post-menopausal breast cancer cases and 194 controls were included in the present study. Circulating levels of BFR and PFAS were measured by gas chromatography coupled to high-resolution mass spectrometry and liquid chromatography coupled to tandem mass spectrometry, respectively. The first statistical approach was based on Principal Component Analysis (PCA) followed by logistic regression models that included the identified principal components as main exposure variables. The second approach used BKMR models with hierarchical variable selection, this latter being suitable for highly correlated exposures. Both approaches were also run separately for Estrogen Receptor positive (ER +) and Estrogen Receptor negative (ER-) breast cancer cases.ResultsPCA identified four principal components accounting for 67% of the total variance. Component 3 showed a marginal association with ER + breast cancer risk. No clear association between BFR and PFAS mixtures and breast cancer was identified using BKMR models, and the credible intervals obtained were very wide. Finally, the BKMR models suggested a negative cumulative effect of BFR and PFAS on ER- breast cancer risk, and a positive cumulative effect on ER + breast cancer risk.ConclusionAlthough globally no clear association was identified, both approaches suggested a differential effect of BFR and PFAS mixtures on ER + and ER- breast cancer risk. However, the results for ER- breast cancer should be interpreted carefully due to the small number of ER- cases included in the study. Further studies evaluating mixtures of substances on larger study populations are needed.

Prediction of Neonatal Respiratory Distress Biomarker Concentration by Application of Machine Learning to Mid-Infrared Spectra.

The authors of this study developed the use of attenuated total reflectance Fourier transform infrared spectroscopy (ATR–FTIR) combined with machine learning as a point-of-care (POC) diagnostic platform, considering neonatal respiratory distress syndrome (nRDS), for which no POC currently exists, as an example. nRDS can be diagnosed by a ratio of less than 2.2 of two nRDS biomarkers, lecithin and sphingomyelin (L/S ratio), and in this study, ATR–FTIR spectra were recorded from L/S ratios of between 1.0 and 3.4, which were generated using purified reagents. The calibration of principal component (PCR) and partial least squares (PLSR) regression models was performed using 155 raw baselined and second derivative spectra prior to predicting the concentration of a further 104 spectra. A three-factor PLSR model of second derivative spectra best predicted L/S ratios across the full range (R2: 0.967; MSE: 0.014). The L/S ratios from 1.0 to 3.4 were predicted with a prediction interval of +0.29, −0.37 when using a second derivative spectra PLSR model and had a mean prediction interval of +0.26, −0.34 around the L/S 2.2 region. These results support the validity of combining ATR–FTIR with machine learning to develop a point-of-care device for detecting and quantifying any biomarker with an interpretable mid-infrared spectrum.

Training load responses modelling and model generalisation in elite sports

This study aims to provide a transferable methodology in the context of sport performance modelling, with a special focus to the generalisation of models. Data were collected from seven elite Short track speed skaters over a three months training period. In order to account for training load accumulation over sessions, cumulative responses to training were modelled by impulse, serial and bi-exponential responses functions. The variable dose-response (DR) model was compared to elastic net (ENET), principal component regression (PCR) and random forest (RF) models, while using cross-validation within a time-series framework. ENET, PCR and RF models were fitted either individually (M_{I}) or on the whole group of athletes (M_{G}). Root mean square error criterion was used to assess performances of models. ENET and PCR models provided a significant greater generalisation ability than the DR model (p = 0.018, p < 0.001, p = 0.004 and p < 0.001 for ENET_{I}, ENET_{G}, PCR_{I} and PCR_{G}, respectively). Only ENET_{G} and RF_{G} were significantly more accurate in prediction than DR (p < 0.001 and p < 0.012). In conclusion, ENET achieved greater generalisation and predictive accuracy performances. Thus, building and evaluating models within a generalisation enhancing procedure is a prerequisite for any predictive modelling.

Determination of LC-PUFAs and Protein Content in Premature infant formula powders by ATR-FTIR Spectroscopy and Chemometrics

Premature infant formula powder (PIFP) contains essential nutrients, such as protein and LC-PUFAs in which inadequate quantities of them may restrict the development of premature infants. The aim of this study was to propose a quantitative analysis for evaluating these nutrients in PIFP. To achieve this purpose, an integrated methodology was developed by combining attenuated total reflectance Fourier-transform infrared spectroscopy (ATR-FTIR) with the chemometrics. ATR-FTIR was utilized to obtain the PIFP spectra in the wavenumber range of 4000-400 cm-1. Principal component analysis (PCA) was employed to perform exploratory analysis of the spectra. Partial least square regression (PLSR) and principal component regression (PCR) models were established for normal spectra, first, and second derivatives to quantitatively determine the protein and LC-PUFAs. The results showed that the PLSR model provided a better prediction than the PCR model with the coefficient of determinations (R2) of 0.992 and 0.983 for training and validation, respectively. It also offers the best prediction for the PLSR model of protein considering the parameters such as R2 (normal spectra) (Cal: 0.995, Val: 0.981). The results revealed that a combination of the ATR-FTIR spectroscopy and the chemometrics models creates a rapid, inexpensive, and non-destructive method for the quantitative analysis of PIFP.

Data-driven self-optimization of processes in the presence of the model-plant mismatch

In this paper, a self-optimization algorithm is developed to find both the optimal operating point and the path from the current condition to the optimal point. Being a model-based strategy, a generalized locally weighted probabilistic principal component regression (PPCR) model that is robust to outliers and can handle missing data, is developed to model the plant. To account for the model-plant mismatch, a penalty term in the form of a robust Gaussian process regression is incorporated into the optimization process. A non-linearity index is utilized to control the accuracy of the local model. Finally, the exploration in optimization for diversity through the acquisition functions is studied. The performance of the proposed algorithm is demonstrated on a simulation case study of a deethanizer column.