K-fold Cross-validation Procedure Research Articles

Taper functions to predict the upper stem diameter of Chir pine (Pinus roxburghii) in the mid-hills of Nepal

A function that predicts the upper stem diameters of an individual tree is helpful in better quantifying the volume and devising sustainable management of commercially and ecologically important species. This study compares three forms of the taper equation (single, variable-exponent, and segmented) using destructive sample data collected from 80 Chir pine (Pinus roxburghii Sarg.) trees across three adjacent stands in the Mid hills of Nepal. A mixed-effects modeling approach was used to evaluate the 13 different taper functions, selecting the best parameter association with the random effect to avoid overparameterization. Additionally, these models were fitted to minimize errors by modeling correlation structure with continuous autoregressive correlation structure of order 1 (corCAR1) and variance with a power variance function. All models were statistically significant at 5% level in the likelihood ratio test when compared with the models fitted without these error minimization. Furthermore, model performance (root mean square error, mean absolute error, and pseudo-R2) was evaluated using k-fold cross-validation procedure. Although all taper functions described more than 95% of the variation in upper stem diameter prediction, the single-form models of Bennett and Swindel (1972) and Amidon (1984), the variable-exponent model form of Sharma and Zhang (2004), and the segmented model form of Max and Burkhart (1976) explained more than 98% percent of the variability. Residual diagnostics indicated that Sharma and Zhang's (2004) model provided better constant and minimum errors in predicting the upper stem diameter along the 8 to 20 m stem length, whereas other models exhibited higher residual errors. Despite model's better performance, variability along the upper stem diameter prediction would affect its applicability in estimating the volume at any merchantable height of individual trees.

Data-driven crack image-based seismic failure mode identification for damaged RC columns

The seismic failure mode identification of reinforced concrete columns is crucial for post-earthquake stability and safety assessment of the damaged concrete buildings. In this paper, a multi-feature crack image-based methodology is proposed for non-destructive non-contact failure mode prediction in reinforced concrete columns damaged in an earthquake. A databank of 455 surface crack patterns from quasi-static experiments on 111 reinforced concrete columns with a variety of structural and geometric properties is collected by the authors aimed at the development and validation of the methodology. For the intricacy retrieval of the surface crack maps, multi-feature fractal-based procedures are taken into consideration. Various machine learning algorithms are employed to train the seismic failure mode predictive models using the characteristics of the damaged RC columns. Ten scenarios are defined based on the image-extracted indices, and structural and geometric characteristics of rectangular reinforced concrete columns. Hyperparameter tuning of the machine learning algorithms using GridsearchCV function has optimized the performance of the models. A k-fold cross-validation procedure is also executed to examine the robustness of the results. Results reveal that the predictive model in which the image-extracted indices and structural parameters are jointly used as input features of supervised learning provides 98% accuracy for the testing dataset with Extreme gradient boost algorithm. In addition, among the scenarios that only image-derived parameters are considered as input features, an accuracy of 83% is obtained with Extreme gradient boost algorithm for the testing dataset.

Development of a Bayesian network-based early warning system for storm-driven coastal erosion

Coastal hazards such as flooding and erosion can cause large economic and human losses. Under this threat, early warning systems can be very cost-effective solutions for disaster preparation. The goal of this study was to develop, test, and implement an operational coastal erosion early warning system supported by a particular method of machine learning. Thus, the system combines Bayesian Networks, and state-of-the-art numerical models, such as XBeach and SWAN, to predict storm erosion impacts in urbanized areas. This system was developed in two phases. In the development phase, all information required to apply the machine learning method was generated including the definition of hundreds of oceanic synthetic storms, modeling of the erosion caused by these storms, and characterization of the impact levels according to a newly defined eerosion iimpact index. This adimensional index relates the distance from the edge of the dune/beach scarp to buildings and the height of that scarp. Finally, a Bayesian Network that acted as a surrogate of the previously generated information was built. After the training of the network, the conditional probability tables were created. These tables constituted the ground knowledge to make the predictions in the second phase. This methodology was validated (1) by comparing 6-h predictions obtained with the Bayesian Network and with process-based models, the latest considered as the benchmark, and (2) by assessing the predictive skills of the Bayesian Network through the unbiased iterative k-fold cross-validation procedure. Regarding the first comparison, the analysis considered the entire duration of three large storms whose return periods were 10, 16, and 25 years, and it was observed that the Bayesian Network correctly predicted between 64% and 72% of the impacts during the course of the storms, depending on the area analyzed. Importantly, this method was also able to identify when the hazardous conditions disappeared after predicting potential consequences. Regarding the Regarding the second validation approach, second validation approach, the k-fold cross-validation procedure was applied to the peak of a set of varying storms and it demonstrated that the predictive skills were maximized (63%–72%) when including three nodes as input conditions of the Bayesian Network. In the operational phase, the system was integrated into the architecture of a forecast and early warning system that predicts emergencies in coastal and port zones in Portugal, and the alerts are issued to authorities every day. This study demonstrated that the two-phase approach developed here can provide fast and high-accuracy predictions of erosion impacts. Also, this methodology can be easily implemented on other sandy beaches constituting a powerful tool for disaster management.

Forecasting project success in the construction industry using adaptive neuro-fuzzy inference system

Project managers often find it a challenge to successfully manage construction projects. As a result, understanding, evaluating, and achieving project success are critical for sponsors to control projects. In practice, determining key success factors and criteria to assess the performance of construction projects and forecast the success of new projects is difficult. To address these concerns, our objective is to go beyond the efficiency-oriented project success criteria by considering both efficiency- and effectiveness-oriented measures to evaluate project success. This paper contributes to existing knowledge by identifying a holistic and multidimensional set of project success factors and criteria using a two-round Delphi technique. We developed a decision support system using the Adaptive Neuro-Fuzzy Inference System (ANFIS) to forecast the success of mid- and large-sized construction projects. We gathered data from 142 project managers in Australia and New Zealand to implement the developed ANFIS. We then validated the constructed ANFIS using the K-fold cross-validation procedure and a real case study of a large construction project in Western Australia. The forecasting accuracy measures R2=0.97461, MAPE = 2.57912%, MAE = 1.88425, RMSE = 2.3610, RRMSE = 0.03149, and PI = 0.01589 suggest that the developed ANFIS is a very good predictor of project success.

Artificial neural network approach for calculating mass attenuation coefficient of different glass systems

In this study, we propose an alternative approach using Artificial Neural Networks (ANN) for determining Mass Attenuation Coefficients (MAC) in various glass systems. This method takes into account the weights of glass compositions, density, and photon energy as input features. The ANN model was trained and tested on a dataset consisting of 650 data points and subsequently validated through a K-fold cross-validation procedure. Our findings demonstrate a high level of accuracy, with R2 values ranging from 0.90 to 0.99. Additionally, the model exhibits robust extrapolation capabilities with an R2 score of 0.87 for predicting MAC values in a new glass system. Furthermore, this approach significantly reduces the need for costly and time-consuming computations and experiments, making it a potential tool for selecting materials for effective radiation protection.

Ridge Regression Under Dense Factor Augmented Models

This article establishes a comprehensive theory of the optimality, robustness, and cross-validation selection consistency for the ridge regression under factor-augmented models with possibly dense idiosyncratic information. Using spectral analysis for random matrices, we show that the ridge regression is asymptotically efficient in capturing both factor and idiosyncratic information by minimizing the limiting predictive loss among the entire class of spectral regularized estimators under large-dimensional factor models and mixed-effects hypothesis. We derive an asymptotically optimal ridge penalty in closed form and prove that a bias-corrected k-fold cross-validation procedure can adaptively select the best ridge penalty in large samples. We extend the theory to the autoregressive models with many exogenous variables and establish a consistent cross-validation procedure using the what-we-called double ridge regression method. Our results allow for nonparametric distributions for, possibly heavy-tailed, martingale difference errors and idiosyncratic random coefficients and adapt to the cross-sectional and temporal dependence structures of the large-dimensional predictors. We demonstrate the performance of our ridge estimators in simulated examples as well as an economic dataset. All the proofs are available in the supplementary materials, which also includes more technical discussions and remarks, extra simulation results, and useful lemmas that may be of independent interest.

Multi-station automatic classification of seismic signatures from the Lascar volcano database

Abstract. This study was aimed to build a multi-station automatic classification system for volcanic seismic signatures such as hybrid, long period, tremor, tectonic, and volcano–tectonic events. This system was based on a probabilistic model made using transfer learning, which has, as the main tool, a pre-trained convolutional network named AlexNet. We designed five experiments using different datasets with data that were real, synthetic, two different combinations of these (combined 1 and combined 2), and a balanced subset without synthetic data. The experiment presented the highest scores when a process of data augmentation was introduced into processing sequence. Thus, the lack of real data in some classes (imbalance) dramatically affected the quality of the results, because the learning step (training) was overfitted to the more numerous classes. To test the model stability with variable inputs, we implemented a k-fold cross-validation procedure. Under this approach, the results reached high predictive performance, considering that only the percentage of recognition of the tectonic events (TC) class was partially affected. The results obtained showed the performance of the probabilistic model, reaching high scores over different test datasets. The most valuable benefit of using this technique was that the use of volcano seismic signals from multiple stations provided a more generalizable model which, in the near future, can be extended to multi-volcano database systems. The impact of this work is significant in the evaluation of hazard and risk by monitoring the dynamic evolution of volcanic centers, which is crucial for understanding the stages in a volcano’s eruptive cycle.

Oxidative Dehydrogenation of Ethane with CO2 over the Fe-Co/Al2O3 Catalyst: Experimental Data Assisted AI Models for Prediction of Ethylene Yield

In this work, Fe-Co-based mixed metal oxides supported on Al2O3 are proposed for ethylene production through oxidative dehydrogenation of ethane with CO2 (ODH-CO2). Thermodynamic feasibility analysis followed by a systematic experimental study is performed on catalyst synthesis and its composition optimization along with process condition optimization in a fixed bed reactor. The study revealed that 5% Fe loaded on 10% Co/Al2O3, 700 °C, and 1:1 are the optimal composition, temperature, and molar ratio of CO2 to ethane, respectively, achieving 29% ethane conversion and resulting in 16% ethylene yield. Further, the experimental data was used to develop different linear, nonlinear, and ensemble AI models for ethylene yield prediction through a systematic grid search and k-fold cross-validation procedure. Among all the models, the kernel ridge regression model is found to be the most accurate, exhibiting the highest R2 value of 0.966 and lowest root mean-squared error (RMSE) of 0.032 on test data, successfully capturing the underlying nonlinear dynamics of ODH-CO2.

DNA and RNA Profiles in Machine Learning Algorithm to Predict Which Patients with AML/MDS Will Respond to Venetoclax-Based Therapy

Introduction Venetoclax in combination with hypomethylation agents is currently used as an alternative to standard induction therapy in patients with acute myeloid leukemia (AML) or advanced myelodysplastic syndrome (MDS). This is particularly relevant when patients are not eligible for standard induction therapy due to age or comorbidity. However, little is known about which patients will benefit from such therapy and which patients will not respond. Selecting the right patient for such therapy is very important considering the clinical fragility of these patients. Genomic profiling using DNA and RNA is becoming a standard of care in clinical testing and there is a need to expand on the clinical utility of the generated data for this profiling. We explored the potential of using these genomic DNA and RNA profiles in a machine learning algorithm to predict which patients with AML/MDS will respond to venetoclax-based therapy. Methods: DNA and RNA from bone marrow samples from 46 patients with AML or advanced high-grade MDS treated with venetoclax (plus either azacitidine or decitabine) were extracted. Using next generation sequencing (NGS), the DNA was sequenced using a 177-gene panel and the RNA was sequenced using a 1408-gene panel. Sequencing was performed using the Illumina NextSeq 550 platform. Ten million reads per sample in a single run were required. Read length was 2 × 150 bp. Expression profiles were generated using Cufflinks. We developed a machine learning algorithm that first selected the relative genes based on performance of each gene with cross-validation and based on stability measures using statistical significance tests. The selected genes were then used to predict response or survival with a k-fold cross-validation procedure (k=12). A naïve Bayesian classifier was constructed on the training of k-1 subsets and tested on the other testing subset. We applied geometric mean naïve Bayesian (GMNB) as the classifier for prediction. Results: Of the 46 patients included in the study, 18 (39%) were females. The median age was 70.5 years (range 33-84). The most common mutations detected in these patients were: ASXL1 (41%), RUNX1 (26%), DNMT3A (24%), FLT3 (21%), NPM1 (20%), NRAS (20%), IDH1/2 (20%), TET2 (17%), TP53 (15%), and SRSF2 (11%). Of these 46 patients, 17 (37%) achieved complete response (CR). Using machine learning we were able to predict those who achieved CR with AUC of 0.972 (95% CI: 0.914-1.00) (Figure), sensitivity of 94.1% and specificity of 93.1%. To achieve this prediction only 10 genes were necessary. There was no significant difference between responders and non-responders in expression level of BCL2 (P=0.11), BAX (P=0.18), or MCL1 (P=0.54). BCL2L1 (BCL-XL) was significantly higher in non-responders (P=0.006) as compared with responders. TP53 was mutated in seven patients, six of whom were non-responders, but no other statistically significant association between specific mutation and response occurred. With a median follow-up of 13 months, 28 (61%) patients were alive and 18 (39%) were dead. Using a machine learning algorithm, we were able to predict overall survival with AUC of 0.970 (95% CI: 0.923-1.00 ) (Figure), sensitivity of 100% and specificity of 94.4%. The algorithm selected 90 genes for this prediction. Conclusion: This data shows that bone marrow expression profiling when used in machine learning can provide a valuable and practical approach for predicting response and overall survival to venetoclax-based therapy in patients with AML/MDS. This prediction requires complex data from a large number of genes (10 to 90 genes) that must be incorporated in a machine learning algorithm. While further studies are needed for confirmation, genomic profiling along with machine learning algorithms can be a practical test for selecting a specific therapeutic approach for specific patients with AML/MDS. Figure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal

Abstract 14239: Development and Validation of a Daily Measure of Cardiorespiratory Fitness - Estimated Vo 2 Max - Using Multiparametric Data From a Wearable Sensor

Introduction: The objective determination of a person’s cardiorespiratory fitness over time, especially in those with heart failure, can play a critical role in optimizing individual therapies, identifying early signs of decompensation, and accelerate the development of novel therapeutic interventions. However, existing methods, such as formal cardiopulmonary exercise testing (CPET), or surrogates like a 6-minute walk test (6MWT) are difficult for both patients and health systems and provide only sparse samples. Hypothesis: Using a wearable ECG-patch sensor and machine learning methods, cardiorespiratory fitness can be estimated daily using routine vital sign dynamics during free-living activities. Methods: CPETs and multiple days of ECG-patch sensor data were collected within 2 weeks of each other for 228 participants (146 normal, and 82 with Heart Failure). A model for VO 2 max estimation was developed using a two-stage procedure. (Figure) In the first stage, the model was pretrained on an unsupervised regression task on over 2000 unique patients’ data, representing over 200 patient-years of activity of daily living. The second stage fine-tuned the pretrained model to predict VO 2 max, using a K-Fold cross validation procedure with 50% train, 25% validation, and 25% testing for four folds. Results: CPET-determined VO 2 max values ranged from 7.1 to 69.1 ml/kg/min (median=28.9, IQR=24.5). The Pearson Correlation Coefficient (PCC) and Mean Absolute Error (MAE) of the estimated VO 2 max was 0.85 and 6.2 ml/kg/min, respectively. For the heart failure subgroup, the PCC and MAE are 0.73 and 3.7 ml/kg/min, respectively. (Figure) Conclusion: Our results show that estimated VO 2 max can be accurately and conveniently determined at high frequency in individuals with a wide range of cardiorespiratory fitness using data from a patch sensor during routine daily activities. Ongoing work in a large HF trial will evaluate its performance over time relative to 6MWT and surveys.

Structural Health Monitoring for impact localisation via machine learning

The focus of this manuscript is Machine learning applied to the Structural Health Monitoring. Two different algorithms have been used to detect impacts on an aluminium plate, one based on the polynomial regression, the other one based on a shallow neural network. Both are supervised algorithms in which some data impacts are used for training, while a complementary subset of data is used for test. The two methods have been preliminarily optimized in terms of training and testing performance and, subsequently, compared in terms of accuracy. By using K-Fold cross validation procedure, with different combinations of training/test sets, the performances of the polynomial models with different degrees were evaluated calculating the Mean Radial Error. For the shallow neural network, three type of learning algorithms were compared: Levenberg-Marquardt, Bayesian Regularization, Scaled Conjugate Gradient. The study confirmed the effectiveness of Machine learning applied to the detection of impacts.

Life Course Digital Twins–Intelligent Monitoring for Early and Continuous Intervention and Prevention (LifeTIME): Proposal for a Retrospective Cohort Study

BackgroundMultimorbidity, which is associated with significant negative outcomes for individuals and health care systems, is increasing in the United Kingdom. However, there is a lack of knowledge about the risk factors (including health, behavior, and environment) for multimorbidity over time. An interdisciplinary approach is essential, as data science, artificial intelligence, and engineering concepts (digital twins) can identify key risk factors throughout the life course, potentially enabling personalized simulation of life-course risk for the development of multimorbidity. Predicting the risk of developing clusters of health conditions before they occur would add clinical value by enabling targeted early preventive interventions, advancing personalized care to improve outcomes, and reducing the burden on health care systems.ObjectiveThis study aims to identify key risk factors that predict multimorbidity throughout the life course by developing an intelligent agent using digital twins so that early interventions can be delivered to improve health outcomes. The objectives of this study are to identify key predictors of lifetime risk of multimorbidity, create a series of simulated computational digital twins that predict risk levels for specific clusters of factors, and test the feasibility of the system.MethodsThis study will use machine learning to develop digital twins by identifying key risk factors throughout the life course that predict the risk of later multimorbidity. The first stage of the development will be the training of a base predictive model. Data from the National Child Development Study, the North West London Integrated Care Record, the Clinical Practice Research Datalink, and Cerner’s Real World Data will be split into subsets for training and validation, which will be done following the k-fold cross-validation procedure and assessed with the Prediction Model Risk of Bias Assessment Tool (PROBAST). In addition, 2 data sets—the Early-Life Data Cross-linkage in Research study and the Children and Young People’s Health Partnership randomized controlled trial—will be used to develop a series of digital twin personas that simulate clusters of factors to predict different risk levels of developing multimorbidity.ResultsThe expected results are a validated model, a series of digital twin personas, and a proof-of-concept assessment.ConclusionsDigital twins could provide an individualized early warning system that predicts the risk of future health conditions and recommends the most effective intervention to minimize that risk. These insights could significantly improve an individual’s quality of life and healthy life expectancy and reduce population-level health burdens.International Registered Report Identifier (IRRID)PRR1-10.2196/35738

Bone Marrow-Based Biomarkers for Predicting aGVHD Using Targeted RNA Next Generation Sequencing and Machine Learning

Bone Marrow-Based Biomarkers for Predicting aGVHD Using Targeted RNA Next Generation Sequencing and Machine Learning

A new framework for experimental design using Bayesian Evidential Learning: The case of wellhead protection area

Groundwater management practices, such as sustainable drinking water extraction or contamination protection, can have significant socioeconomic impacts. A complete uncertainty analysis should ideally be performed to anticipate all possible outcomes and assess any risk. Uncertainties arise as a result of our limited understanding of the physical processes involved, as well as a scarcity of measurement data, whether directly or indirectly related to the physical parameters of interest. In this paper, we use a small number of tracing experiments (predictor) to predict the wellhead protection area (WHPA, target), the shape and extent of which are influenced by the distribution of hydraulic conductivity (K). Our first goal is to make stochastic predictions of the WHPA within the Bayesian Evidential Learning (BEL) framework, which uses machine learning to find a direct relationship between predictor and target. This relationship is learned using a small number of training models (400) drawn from the prior distribution of K. Forward modelling is used to obtain the 400 pairs of simulated predictors and targets. Newly collected field data can then be used directly to predict the approximate posterior distribution of the corresponding WHPA, obviating the need for the traditional step of data inversion. The number and location of data sources (injection wells) influence the posterior WHPA distribution's uncertainty range. Our second goal is to extend BEL to determine the optimal design of data source locations that minimises the WHPA's posterior uncertainty. Because the BEL model, once trained, allows the computation of the posterior uncertainty corresponding to any new input data, experimental design can be done explicitly, without averaging or approximating. We estimate the WHPA's posterior uncertainty range using the Modified Hausdorff Distance (MHD) and Structural Similarity (SSIM) index metrics. Because the breakthrough curves store information on a large area of the K field surrounding the pumping well, increasing the number of injection wells reduces the derived posterior WHPA uncertainty. Our method can also estimate which injection wells are more informative than others, as demonstrated by a k-fold cross-validation procedure. Overall, the application of BEL to experimental design allows for identifying data sources that maximise the information content of any measurement data while keeping budget constraints and computational costs to a minimum.

A Machine Learning Approach to Determine Airport Asphalt Concrete Layer Moduli Using Heavy Weight Deflectometer Data

An integrated approach based on machine learning and data augmentation techniques has been developed in order to predict the stiffness modulus of the asphalt concrete layer of an airport runway, from data acquired with a heavy weight deflectometer (HWD). The predictive model relies on a shallow neural network (SNN) trained with the results of a backcalculation, by means of a data augmentation method and can produce estimations of the stiffness modulus even at runway points not yet sampled. The Bayesian regularization algorithm was used for training of the feedforward backpropagation SNN, and a k-fold cross-validation procedure was implemented for a fair performance evaluation. The testing phase result concerning the stiffness modulus prediction was characterized by a coefficient of correlation equal to 0.9864 demonstrating that the proposed neural approach is fully reliable for performance evaluation of airfield pavements or any other paved area. Such a performance prediction model can play a crucial role in airport pavement management systems (APMS), allowing the maintenance budget to be optimized.

Cross-validation strategies in QSPR modelling of chemical reactions

ABSTRACT In this article, we consider cross-validation of the quantitative structure-property relationship models for reactions and show that the conventional k-fold cross-validation (CV) procedure gives an ‘optimistically’ biased assessment of prediction performance. To address this issue, we suggest two strategies of model cross-validation, ‘transformation-out’ CV, and ‘solvent-out’ CV. Unlike the conventional k-fold cross-validation approach that does not consider the nature of objects, the proposed procedures provide an unbiased estimation of the predictive performance of the models for novel types of structural transformations in chemical reactions and reactions going under new conditions. Both the suggested strategies have been applied to predict the rate constants of bimolecular elimination and nucleophilic substitution reactions, and Diels-Alder cycloaddition. All suggested cross-validation methodologies and tutorial are implemented in the open-source software package CIMtools (https://github.com/cimm-kzn/CIMtools).

Discrimination of Sweet Cherry Cultivars Based on Electronic Tongue Potentiometric Fingerprints

Sweet cherry is highly appreciated by its characteristic flavor, which conditions the consumer’s preference. In this study, four sweet cherry cultivars (Durona, Lapins, Summit, and Van cultivars) were characterized according to biometric (fruit and stone weights, length, maximum and minimum diameters, pulp/stone mass ratio), physicochemical (CIELAB color, penetration force, titratable acidity, and total soluble solids), and potentiometric profiles (recorded by a lab-made electronic tongue with lipid polymeric membranes). Biometric and physicochemical data were significantly cultivar-dependent (p-value < 0.0001, one-way ANOVA). Summit cherries had higher masses and dimensions. Lapins cherries had the highest penetration force values having, together with Summit cherries, the highest CIELAB values. Van cherries showed the highest total soluble solids contents. No significant differences were found for fruits’ acidity (similar titratable acidities). The possibility of discriminating cherry cultivars was also evaluated using a linear discriminant analysis/simulated-annealing algorithm. A discriminant model was established based on nine non-redundant biometric-physicochemical parameters (using a low-level data fusion), with low sensitivity (75 ± 15% for the repeated K-fold cross-validation). On the contrary, a discriminant model, based on the potentiometric fingerprints of 11 selected sensors, allowed a better discrimination, with sensitivities of 88 ± 7% for the repeated K-fold cross-validation procedure. Thus, the electronic tongue could be used as a practical tool to discriminate cherry cultivars and, if applied by fruit traders, may reduce the risk of mislabeling, increasing the consumers’ confidence when purchasing this high-value product.

Sweet peppers discrimination according to agronomic production mode and maturation stage using a chemical-sensory approach and an electronic tongue

The demand for organic foods has increased worldwide, in particular due to the association with healthier, more nutritious and tasty products, being a clear trend on sweet peppers’ consumption. Thus, this study aimed to evaluate the effects of agronomic production mode (conventional and organic) and maturation stage (associated to green, turning and red colours) on the chemical-sensory attributes of peppers grown in open field. It was found that organic peppers had a better visual/tactile aspect (greater firmness and more intense colours) but lower chemical quality (lower titratable acidity and total soluble solids). On the other hand, red peppers (higher maturation stage) had lower visual-tactile quality but higher chemical quality. From sensory analysis, conventional peppers had better overall aspect, colour intensity-homogeneity and brightness. Then again, the maturation stage of peppers mostly influenced the sensory visual attributes, being turning colour peppers the less appreciated, although organic red peppers were less succulent and had a lower global quality. Even so, the chemical-sensory parameters could be used to discriminate peppers taking into account the agronomic production mode and the maturation stage/colour (79 ± 12% of correct classifications for the repeated K-fold cross-validation procedure). However, a trained sensory panel is required, which can be a major drawback considering their scarcity. This limitation was successfully overcome by using a potentiometric electronic tongue, which allowed discriminating the peppers with a higher predictive sensitivity (85 ± 9%), showing that this device could be used as an accurate taste sensor for the qualitative analysis of sweet peppers.

A cascaded deep convolution neural network based CADx system for psoriasis lesion segmentation and severity assessment

The design of an efficient computer-aided diagnosis (CADx) system for psoriasis severity assessment demands both accurate segmentation and classification of psoriasis lesions. Recently, few studies have been conducted to design automatic CADx systems for psoriasis severity assessment using traditional machine learning approaches. However, these approaches are highly featured dependent and require extensive and careful feature extraction. Among a large number of features extracted, assessing the features which contribute significantly to the classifier performing is a difficult and time-consuming task. Large features lead to poor generalization, due to high inter and intra-class variation of psoriasis skin lesions. This makes the task of implementing a reliable CADx system challenging. In such similar cases, Deep learning-based approaches have been proven better because of their ability to learn and make intelligent decisions automatically. In this study, a fully automated deep learning-based CADx system for psoriasis has been proposed. The system combines three modules in a single framework for achieving different objectives namely; recognition of psoriasis and non-psoriasis disease, automatic segmentation of psoriatic lesion, and its severity assessment. The modified U-Net and modified VGG-16 model have been implemented and trained for the segmentation and classification task respectively. The severity assessment module is capable of extracting discriminative features specifically related to the psoriatic lesion, which is automatically segmented by the segmentation module. The performance of the proposed CADx framework has been extensively evaluated on an extensive psoriasis dataset using k-fold cross-validation procedure. The appropriateness of the proposed system has been justified in terms of its performance at each of the three stages along with benchmarking against previously reported systems. Further, the system accuracy and reliability index has been evaluated for a dataset of varying size to validate the consistency of the proposed system.

Factor-Adjusted Ridge Prediction Using Large-Dimensional Mixed-Effects Models

This paper establishes a comprehensive theory of the optimality, robustness, and cross-validation selection-consistency for the ridge regression under factor-augmented models with possibly dense idiosyncratic information. Using spectral analysis for random matrices, we show that the ridge regression is asymptotically efficient in capturing both factor and idiosyncratic information, by minimizing the limiting predictive loss among the entire class of spectral regularized estimators under large-dimensional factor models and mixed-effects hypothesis. We derive an asymptotically optimal ridge penalty in closed form, and prove that a bias-corrected k-fold cross-validation procedure can select the best ridge penalty adaptively in large samples. We extend the theory to the autoregressive models with many exogenous variables and establish a consistent cross-validation procedure using the what-we-called double ridge regression method. Our results allow for non-parametric distributions for martingale difference errors and idiosyncratic random coefficients, and adapt to both the cross-sectional and temporal dependence structures of the large-dimensional predictors. A simulation study shows the efficiency of the ridge regression estimators under dense models and its robustness against sparsity. We apply our methods to forecast the growth rate of US industrial production and S\&P 500 index using the FRED-MD database. Our double ridge regression improves the autoregressive prediction and outperforms the principal component regression and the factor-adjusted LASSO regression consistently.