Variable Selection Algorithm Research Articles

The Improved EMS Algorithm for Latent Variable Selection in M3PL Model

One of the main concerns in multidimensional item response theory (MIRT) is to detect the relationship between items and latent traits, which can be treated as a latent variable selection problem. An attractive method for latent variable selection in multidimensional 2-parameter logistic (M2PL) model is to minimize the observed Bayesian information criterion (BIC) by the expectation model selection (EMS) algorithm. The EMS algorithm extends the EM algorithm and allows the updates of the model (e.g., the loading structure in MIRT) in the iterations along with the parameters under the model. As an extension of the M2PL model, the multidimensional 3-parameter logistic (M3PL) model introduces an additional guessing parameter which makes the latent variable selection more challenging. In this paper, a well-designed EMS algorithm, named improved EMS (IEMS), is proposed to accurately and efficiently detect the underlying true loading structure in the M3PL model, which also works for the M2PL model. In simulation studies, we compare the IEMS algorithm with several state-of-art methods and the IEMS is of competitiveness in terms of model recovery, estimation precision, and computational efficiency. The IEMS algorithm is illustrated by its application to two real data sets.

Improving Raman-Based Models for Real-Time Monitoring the CHO Cell Culture Process with Effective Variable Selection Strategies

Research has shown that Raman spectroscopy could be applied to monitor various components in mammalian cell culture in real time. In the process of application, it is necessary to ensure the performance of the Raman-based model. The variable selection strategy is an effective method that significantly influences the model performance and simplification. In this study, different variable selection strategies were evaluated, and the optimal variable selection strategy was determined for monitoring the CHO cell culture process. Firstly, a wide variety of spectral regions involving the Raman fingerprinting region and the C-H stretching region were investigated. Secondly, six different variable selection algorithms were meticulously assessed. Thirdly, the combination of different variable selection algorithms was used to improve model performance and simplify the model. Finally, the monitoring of cell culture processes was implemented. The findings underscored that commonly used spectral regions could improve the model performance but could not simplify the model well. Moving-window partial least square (MWPLS), genetic algorithm (GA), and random frog (RF) are more suitable for Raman modeling of the cell culture process, but they must be used after the spectral region selection. The combination of three variable selection algorithms (MWPLS-GA-RF) improved the model’s performance by 16–70% by selecting 30–60 variables, effectively simplifying the model. For glucose, lactate, viable cell density, and ammonium ion, real-time monitoring was performed well. This study will be helpful for researchers to select suitable variable selection strategies for building models for the real-time monitoring of cell culture.

Do commodity prices matter for global systemic risk? Evidence from ML variable selection

We identify robust predictors of global systemic risk proxied by conditional capital shortfall (SRISK) among a comprehensive set of commodity prices for the period between January 2004 and December 2021. The search is based on a battery of ML variable selection algorithms which apply both to price levels and price shocks in the presence of control variables, including the first lag of SRISK, world industrial production, global economic policy uncertainty, geopolitical risk as well as the global stance of monetary and macroprudential policies. We find that these controls outweigh commodity prices as the predictors of global systemic risk. Of the commodities themselves, the prices for agricultural commodities, including food, e.g. chicken, bananas, beef, tea, cocoa, are more important predictors of global systemic risk than the prices for energy commodities, e.g. natural gas and oil prices. The financialization of agricultural commodities, bio-energy expansion as well as commodity-specific dependence of the major economies contributing to global systemic risk, e.g. China, account for our main finding. We also document the positive linkage between commodity prices and systemic risk for the majority of commodities. Thus, monitoring commodity prices to avoid their unbalanced growth is of vast importance to curb global systemic financial risk.

Non-Destructive Identification of Wool and Cashmere Fibers Based on Cascade Optimizations of Interval-Wavelength Selection Using NIR Spectroscopy

ABSTRACT Near-infrared (NIR) spectroscopy is an effective method for identifying wool and cashmere fibers, with high spectral data providing a wealth of information. However, a key issue is that the accuracy and robustness of subsequent estimates can be reduced by redundant and interfering wavelengths. For this reason, a novel interval-wavelength cascaded optimization method is proposed. Initially, the collected spectral data are preprocessed by standard normal variate transformation (SNV) to eliminate the scattering effect. Then, the backward interval partial least squares (BiPLS) algorithm is applied for the preliminary selection of spectral intervals, followed by the application of three different variable selection algorithms, competitive adaptive reweighted sampling (CARS), successive projection algorithm (SPA) and whale optimization algorithm (WOA), for secondary wavelength optimization, respectively. Finally, both support vector machine (SVM) and random forest (RF) discriminant models are built to identify the extracted subset of wavelengths. In the experimental stage, the cascade method BiPLS-WOA selects 36 wavelengths, in SVM, the accuracy of the validation set reaches 96.9%, and the area under the ROC curve (AUC) can reach 99.3%. The results demonstrate that the proposed method can eliminate redundant and collinear variables, thereby validating the effectiveness of distinguishing wool and cashmere fibers.

A novel variable selection algorithm based on neural network for near-infrared spectral modeling

BackgroundPartial least squares (PLS) is a widely used technique for modeling spectral data. Researchers have developed numerous PLS-based variable selection algorithms to enhance model predictive ability and interpretability. In recent years, as neural network technology has advanced, these algorithms have been increasingly applied to spectral data modeling. However, current research on neural network modeling tends to prioritize network structure over variable selection. ResultsOur study introduces a neural network-based variable selection algorithm called VSNN (Variable Selection based on Neural Network). By iteratively eliminating unimportant variables using an exponentially decreasing function (EDF), the algorithm achieves the selection of variables in spectral data. VSNN can easily integrate different types of neural networks. In this study, we analyzed the impact of neural network types, activation functions, and variable importance vectors on algorithm performance. We tested the algorithm on four datasets: corn moisture, corn oil, tablets, and meat. The results indicate that VSNN significantly enhances the predictive ability of the model compared to partial least squares (PLS), neural networks (NN), and Joint Mutual Information Maximisation (JMIM). Specifically, non-linear activation functions markedly improve performance on non-linear meat datasets. Compared to PLS, the Root Mean Square Error of Prediction (RMSEP) values for the four datasets—corn moisture, corn oil, tablets, and meat—decreased from 0.0409, 0.0728, 3.97, and 3.2 to 0.002, 0.0236, 3.12, and 0.36, respectively, after applying the VSNN variable selection algorithm. SignificanceVSNN can serve as a versatile framework to enhance variable selection, modeling, and predictive performance by adapting neural network types and variable importance evaluation indicators. As machine learning technology advances, the strength of VSNN is poised to increase. This study highlights the potential of VSNN as an effective algorithmic framework for variable selection in spectroscopy applications.

Deep learning and feature reconstruction assisted vis-NIR calibration method for on-line monitoring of key growth indicators during kombucha production

Artificial intelligence (AI) technology is advancing the digitization and intelligence development of the food industry. A promising application is using deep learning-assisted visible near-infrared (vis-NIR) spectroscopy to monitor residual sugar and bacterial concentration in real-time, ensuring kombucha quality during production. The feature fingerprints of residual sugar and bacterial concentration were extracted by four variable selection algorithms and then reconstructed using serial and parallel processing methods. Based on these reconstructed features, Partial Least Squares (PLS) and Convolutional Neural Networks (1DCNN and 2DCNN) models were developed and compared. The experimental results showed that the 2DCNN model based on reconstruction features achieved superior performance. The RPDs of the residual sugar and bacterial concentrations models were 4.49 and 6.88, while the MAEs were 0.42 mg/mL and 0.04 (Abs), respectively. These results suggest that the proposed modeling strategy effectively supports quality control during kombucha production and provides a new perspective for spectral analysis.

Feasibility study on non-destructive detection of microplastic content in flour based on portable Raman spectroscopy system combined with mixed variable selection method

Microplastics, as emerging environmental pollutants, have garnered considerable attention due to their contamination of both the environment and food. Microplastics can infiltrate the human food chain through multiple pathways, potentially posing health risks to humans. Currently, non-destructive testing of microplastics in food is considered challenging. This study aims to investigate the feasibility of employing a portable Raman spectroscopy system for non-destructive detection of microplastic content (polystyrene, PS; polyethylene, PE) in flour. In this study, a portable spectrometer was used to collect flour spectra of different abundances of microplastics. To enhance the predictive performance of the partial least squares (PLS) model, a mixed variable selection strategy that combined the wavelength interval selection method (Synergy interval partial least squares, siPLS) and the wavelength point selection method (Least absolute shrinkage and selection operator, LASSO; Multiple feature-spaces ensemble by least absolute shrinkage and selection operator, MFE-LASSO) was proposed. Four regression models (PLS, siPLS, siPLS-LASSO, siPLS-MFE-LASSO) were developed and compared for detecting PS and PE content in flour. The siPLS-MFE-LASSO model exhibited the best generalization performance in the prediction set, and was considered to have the best generalization performance (PS: RP2 = 0.9889, RMSEP=0.0344 %; PE: RP2 = 0.9878, RMSEP=0.0361 %). In conclusion, this study has demonstrated the potential of using a portable Raman spectrometer in conjunction with a mixed variable selection algorithm for non-destructive detection of PS and PE content in flour, providing more possibilities for non-destructive detection of microplastic content in food.

Rapid UV-Vis spectrophotometric method aided by firefly-PLS models for the simultaneous quantification of ciprofloxacin, lomefloxacin, and enrofloxacin in their laboratory mixture, dosage forms and water samples: greenness and blueness assessment

Herein, a novel UV spectrophotometric method coupled with chemometric tools was developed for the simultaneous determination of three fluoroquinolone antibiotics: ciprofloxacin, lomefloxacin, and enrofloxacin. Such integration of UV spectroscopy and chemometric analysis proved to be a simple, rapid, and cost-effective approach for the quantification of these clinically important pharmaceutical compounds and aid in their quality control analysis. The method employed firefly algorithm for variable selection and partial least squares (PLS) regression for model calibration. The developed method was validated by independent test set in addition the accuracy, intra and inter-day precision as per ICH guidelines which showed a satisfactory performance with mean recovery ranged between 98.18 and 101.83 with %RSD < 2. Besides, the developed method displayed ultrasensitive levels with LODs (0.0803, 0.1125, 0.1309 µg/mL) and LOQs (0.2434, 0.3409, 0.3968 µg/mL) for ciprofloxacin, lomefloxacin, and enrofloxacin, respectively. The greenness and blueness of the developed method were also evaluated using the recently proposed Analytical GREEnness metric approach (AGREE) and Blue applicability grade index (BAGI) tools, which showed a high AGREE score of 0.79 and a BAGI score of 77.5. These results indicate that the developed method provides an environmentally friendly alternative to the traditionally used chromatographic techniques, while maintaining high analytical practicability. Finally, the application of the developed methodology was demonstrated on real pharmaceutical and tap water samples, and the results were in good agreement with those obtained by the reference HPLC method indicating the reliability and suitability of the proposed spectrophotometric method for routine analysis of fluoroquinolone antibiotics.

Short wave-near infrared spectroscopy for predicting soluble solid content in intact mango with variable selection algorithms and chemometric model

Mango sweetness is one of the most prominent internal quality constituents that attract consumer's attention. The current common methods used to determine the sweetness of mango have significant disadvantages in that labor-intensive, time-consuming, and damaging techniques. This research aims to predict sweetness in mangoes using short wave-near infrared (SW-NIR) spectroscopy in the ranges 900–1650 nm, along with variable selection algorithms and chemometric model. A total of 120 mango samples were used to collect the spectra using a fibre module SW-NIR spectrometer. The partial least squares (PLS) regression with several spectral preprocessing methods was employed to develop the calibration model, and the best preprocessing technique was selected. The Savitzky–Golay second-derivative preprocessing technique performed better among the other preprocessing techniques with a correlation coefficient of prediction (rpred) of 0.74 and standard error of prediction (SEP) is 0.78 %Brix. After that, two variable selection techniques were used to select effective wavelength variables, including regression coefficient and successive projections algorithm (SPA). For SSC prediction in the range 900–1650 nm, the SPA-PLS model obtained a rpred of 0.78 and SEP of 0.67 %Brix. The current study unequivocally shows that the proposed SW-NIR spectroscopy coupled with a suitable chemometrics method can evaluate mango sweetness nondestructively.

Forest above-ground biomass estimation based on strongly collinear variables derived from airborne laser scanning data

Airborne laser scanning technique (ALS) is the most appealing remote sensing technique for the precise estimation of forest above-ground biomass (AGB). Significantly strong correlations (collinearity) among the independent variables derived from ALS data decrease the accuracy of developed AGB models. To address this issue, we propose a novel variable selection algorithm-an improved sure independence screening (SPV), which integrates the Pearson correlation coefficient, threshold μ, and variance inflation factor. We further compared the performance of SPV-based, stepwise feature selection (SFS)-based, and least absolute shrinkage and selection operator (LASSO)-based AGB models developed with different regression approaches that were sensitive to strong collinearity. Field-measured data and corresponding ALS data, acquired from 1002 sample plots distributed across four distinct forest types within the Guangxi Zhuang Autonomous Region in Southern China, were used to evaluate variable selection techniques and develop AGB models. Results indicated that ALS variables selected by SPV exhibited weaker collinearity compared to those selected by SFS and LASSO. SPV-based AGB models outperformed SFS-based AGB models with higher leave-one-out cross-validation adjusted R2 (LOOCV Radj2) by 0.1% − 27.8%. SPV-based AGB models outperformed LASSO-based AGB models with higher LOOCV Radj2 by 0.4% − 16.3%. Hence, for variable selection in constructing AGB models (linear regression model, log–log regression model, and generalized additive model (GAM)) based on strongly collinear ALS variables, SPV is most preferred, followed by SFS and LASSO. The smooth curves from our GAMs developed using SPV-selected variables revealed that five canopy height variables (whp40, hmean, hp25, hp70, hskew), one canopy density variable (whp85), three density-related variables (nVegPts, bcc, and mcc), and one vertical structural variable (LADmean) were positively correlated with AGB. The canopy height variables (whp40, hmean, hp25, and hp70) were identified as the most important variables in estimating AGB for four forest types. The canopy density variable wdp85 showed a strong effect on AGB of the coniferous forests, whereas it had almost no effect on the AGB of broadleaved forests. Overall, this manuscript proposes a novel variable selection algorithm named SPV, aimed at addressing collinearity of variables derived from ALS data, which has significant implications for the application of ALS in forest inventory and forest modeling.

A detection method of oil content for maize kernels based on CARS feature selection and deep sparse autoencoder feature extraction

To achieve accurate, rapid, and non-destructive oil content determination in maize, a neural fitting model that combines feature selection and feature extraction is proposed. Competitive adaptive re-weighted sampling (CARS) is used to choose features, while the deep sparse autoencoder (DSAE) network is utilized to extract features. The 160 spectral data from the near-infrared spectra public datasets of maize served as experimental samples. Several preprocessing approaches were tested, and the first-order derivatives fared the best. To capture the various degrees of abstract features in spectral data, SSAE networks with two or three hidden layers are developed. The number of nodes in the network's hidden layer (30, 50, 70, 100, 200, 300, respectively) and the network's sparsity parameter (0.001, 0.004, 0.007, 0.01, 0.04, 0.07, 0.1, 0.2, …, 1) are tuned to generate partial least squares regression (PLSR) models for testing the feature learning effect. The highest performance comes from a three-layer network with 200 hidden layer nodes and a sparsity parameter of 0.04. This network extracts three levels of features from the preprocessed spectral data: F1, F2, and F3. Full_cars, F1_cars, F2_cars, and F3_cars are obtained using the CARS algorithm for variable selection on the preprocessed spectral data, F1, F2, and F3, respectively. They are then integrated into a final feature set to build a neural-fitting network (FNN). After training and testing, the test set has a correlation coefficient of 0.964205 and an mean square error (MSE) of 0.00636809. Comparing this model to support vector machine regression (SVR), PLSR, and gaussian process regression (GPR), the findings demonstrate that the model described in this study outperforms the others in terms of fitting accuracy and generalization performance. To summarize, this paper introduces a new regression analysis method based on near-infrared spectra.

Development of a diagnostic predictive model for determining child stunting in Malawi: a comparative analysis of variable selection approaches

BackgroundChildhood stunting is a major indicator of child malnutrition and a focus area of Global Nutrition Targets for 2025 and Sustainable Development Goals. Risk factors for childhood stunting are well studied and well known and could be used in a risk prediction model for assessing whether a child is stunted or not. However, the selection of child stunting predictor variables is a critical step in the development and performance of any such prediction model. This paper compares the performance of child stunting diagnostic predictive models based on predictor variables selected using a set of variable selection methods.MethodsFirstly, we conducted a subjective review of the literature to identify determinants of child stunting in Sub-Saharan Africa. Secondly, a multivariate logistic regression model of child stunting was fitted using the identified predictors on stunting data among children aged 0–59 months in the Malawi Demographic Health Survey (MDHS 2015–16) data. Thirdly, several reduced multivariable logistic regression models were fitted depending on the predictor variables selected using seven variable selection algorithms, namely backward, forward, stepwise, random forest, Least Absolute Shrinkage and Selection Operator (LASSO), and judgmental. Lastly, for each reduced model, a diagnostic predictive model for the childhood stunting risk score, defined as the child propensity score based on derived coefficients, was calculated for each child. The prediction risk models were assessed using discrimination measures, including area under-receiver operator curve (AUROC), sensitivity and specificity.ResultsThe review identified 68 predictor variables of child stunting, of which 27 were available in the MDHS 2016–16 data. The common risk factors selected by all the variable selection models include household wealth index, age of the child, household size, type of birth (singleton/multiple births), and birth weight. The best cut-off point on the child stunting risk prediction model was 0.37 based on risk factors determined by the judgmental variable selection method. The model’s accuracy was estimated with an AUROC value of 64% (95% CI: 60%-67%) in the test data. For children residing in urban areas, the corresponding AUROC was AUC = 67% (95% CI: 58–76%), as opposed to those in rural areas, AUC = 63% (95% CI: 59–67%).ConclusionThe derived child stunting diagnostic prediction model could be useful as a first screening tool to identify children more likely to be stunted. The identified children could then receive necessary nutritional interventions.

Enhancing the streamflow simulation of a process-based hydrological model using machine learning and multi-source data

Streamflow simulation is crucial for flood mitigation, ecological protection, and water resource planning. Process-based hydrological models and machine learning algorithms are the mainstream tools for streamflow simulation. However, their inherent limitations, such as time-consuming and large data requirements, make achieving high-precision simulations challenging. This study developed a hybrid approach to simultaneously improve the accuracy and computational efficiency of streamflow simulation, which integrates Block-wise use of the TOPMODEL (BTOP) model into the eXtreme Gradient Boosting (XGBoost), i.e., BTOP_XGB. In this approach, BTOP generates simulated streamflow using the Latin hypercube sampling algorithm instead of the time-consuming calibration algorithms to reduce computational costs. Then, XGBoost combines BTOP simulated streamflow with multi-source data to reduce simulation errors. In which, serval input variable selection algorithms are employed to choose relevant inputs and remove redundant information for model. The hybrid approach is validated and compared with a standalone model at three hydrological stations in the Jialing River basin, China. The results show that the performance of BTOP_XGB is significantly better than the BTOP and XGBoost models. The NSE of BTOP_XGB at Beibei, Xiaoheba, and Luoduxi stations increases by 54%, 21%, and 83%, respectively. Meanwhile, the computational time of BTOP_XGB is saved by >90% compared to the original calibrated BTOP. BTOP_XGB is less affected by parameter sample sizes and data amounts, demonstrating the robustness of the hybrid model. This study simplifies the complexity of the hydrological model and enhances the stability of machine learning, jointly improving the reliability of streamflow simulation. The hybrid approach provides a potential shortcut for streamflow simulation over basins with large areas or limited observed data.

In-house validation of a visible and near infrared spectroscopy non-targeted method to support panel test of virgin olive oils

In this study, an in-house validation of Visible and Near Infrared Spectroscopy was performed to distinguish between extra virgin olive oil (EVOO) and virgin olive oil (VOO). A total of 161 samples of olive oil of three different categories (EVOO, VOO and lampante (LOO)) were analysed by transflectance using a monochromator instrument. One-class models were initially developed using Partial Least Squares (PLS) Density Modelling to characterize EVOO and VOO category. Once the LOO samples were discriminated, linear and non-linear discriminant models were built to classify EVOO and VOO. Different data pre-treatments and variable selection algorithms were evaluated to establish the best models in terms of Correct Classification Rate (CCR). The best model, obtained after variable selection using PLS Discriminant Analysis, yielded CCR values of 82.35 % for EVOO and 66.67 % for VOO in external validation. These results confirmed that VIS + NIRS technology may be used to provide rapid, non-destructive preliminary screening of olive oil samples for categorization; suspect samples may then be analysed by official analytical methods.

Versatile Descent Algorithms for Group Regularization and Variable Selection in Generalized Linear Models

This article proposes an adaptively bounded gradient descent (ABGD) algorithm for group elastic net penalized regression. Unlike previously proposed algorithms, the proposed algorithm adaptively bounds the Fisher information matrix, which results in a flexible and stable computational framework. In particular, the proposed algorithm (i) does not require orthogonalization of the predictors, and (ii) can be easily applied to any combination of exponential family response distribution and link function. The proposed algorithm is implemented in the grpnet R package (available from CRAN), which implements the approach for common response distributions (Gaussian, binomial, and Poisson), as well as several response distributions not previously considered in the group penalization literature (i.e., multinomial, negative binomial, gamma, and inverse Gaussian). Simulated and real data examples demonstrate that the proposed algorithm is as or more efficient than existing methods for Gaussian, binomial, and Poisson distributions. Furthermore, using two genomic examples, I demonstrate how the proposed algorithm can be applied to high-dimensional multinomial regression problems with grouped predictors. R code to reproduce the results is included as supplementary materials.

Near-infrared spectroscopy and multivariate analysis as effective, fast, and cost-effective methods to discriminate Candida auris from Candida haemulonii.

Candida auris and Candida haemulonii are two emerging opportunistic pathogens that have caused an increase in clinical cases in the recent years worldwide. The differentiation of some Candida species is highly laborious, difficult, costly, and time-consuming depending on the similarity between the species. Thus, this study aimed to develop a new, faster, and less expensive methodology for differentiating between C. auris and C. haemulonii based on near-infrared (NIR) spectroscopy and multivariate analysis. C. auris CBS10913 and C. haemulonii CH02 were separated in 15 plates per species, and three isolated colonies of each plate were selected for Fourier transform near-infrared (FT-NIR) analysis, totaling 90 spectra. Subsequently, principal component analysis (PCA) and variable selection algorithms, including the successive projections algorithm (SPA) and genetic algorithm (GA) coupled with linear discriminant analysis (LDA), were employed to discern distinctive patterns among the samples. The use of PCA, SPA, and GA algorithms associated with LDA achieved 100% sensitivity and specificity for the discriminations. The SPA-LDA and GA-LDA algorithms were essential in selecting the variables (infrared wavelengths) of most importance for the models, which could be attributed to binding of cell wall structures such as polysaccharides, peptides, proteins, or molecules resulting from yeasts' metabolism. These results show the high potential of combined FT-NIR and multivariate analysis techniques for the classification of Candida-like fungi, which can contribute to faster and more effective diagnosis and treatment of patients affected by these microorganisms.

Prediction of cyanidin 3-rutinoside content in Michelia crassipes based on near-infrared spectroscopic techniques.

Currently the determination of cyanidin 3-rutinoside content in plant petals usually requires chemical assays or high performance liquid chromatography (HPLC), which are time-consuming and laborious. In this study, we aimed to develop a low-cost, high-throughput method to predict cyanidin 3-rutinoside content, and developed a cyanidin 3-rutinoside prediction model using near-infrared (NIR) spectroscopy combined with partial least squares regression (PLSR). We collected spectral data from Michelia crassipes (Magnoliaceae) tepals and used five different preprocessing methods and four variable selection algorithms to calibrate the PLSR model to determine the best prediction model. The results showed that (1) the PLSR model built by combining the blockScale (BS) preprocessing method and the Significance multivariate correlation (sMC) algorithm performed the best; (2) The model has a reliable prediction ability, with a coefficient of determination (R2) of 0.72, a root mean square error (RMSE) of 1.04%, and a residual prediction deviation (RPD) of 2.06. The model can be effectively used to predict the cyanidin 3-rutinoside content of the perianth slices of M. crassipes, providing an efficient method for the rapid determination of cyanidin 3-rutinoside content.

Detection of protein content in alfalfa using visible/ near-infrared spectroscopy technology

In this study, a quantitative model was developed using near-infrared spectroscopy to analyze protein content in dried purple alfalfa, employing preprocessing methods (SG, SNV, MSC, FD) and variable selection algorithms (CARS, IRIV) to optimize spectra. Models using ELM, PLSR, SVM, and LSTM were tested; the MSC-CARS-PLSR-SVM model achieved the highest accuracy, with a calibration determination coefficient (R²) of 0.9982 and root mean square error (RMSE) of 0.1088, and a prediction R² of 0.9645 with RMSE of 0.5230, offering a precise and reliable method for protein content prediction.

Evaluation of a machine-learning model based on laboratory parameters for the prediction of acute leukaemia subtypes: a multicentre model development and validation study in France

Acute leukaemias are life-threatening haematological cancers characterised by the infiltration of transformed immature haematopoietic cells in the blood and bone marrow. Prompt and accurate diagnosis of the three main acute leukaemia subtypes (ie acute lymphocytic leukaemia [ALL], acute myeloid leukaemia [AML], and acute promyelocytic leukaemia [APL]) is of utmost importance to guide initial treatment and prevent early mortality but requires cytological expertise that is not always available. We aimed to benchmark different machine-learning strategies using a custom variable selection algorithm to propose an extreme gradient boosting model to predict leukaemia subtypes on the basis of routine laboratory parameters. This multicentre model development and validation study was conducted with data from six independent French university hospital databases. Patients aged 18 years or older diagnosed with AML, APL, or ALL in any one of these six hospital databases between March 1, 2012, and Dec 31, 2021, were recruited. 22 routine parameters were collected at the time of initial disease evaluation; variables with more than 25% of missing values in two datasets were not used for model training, leading to the final inclusion of 19 parameters. The performances of the final model were evaluated on internal testing and external validation sets with area under the receiver operating characteristic curves (AUCs), and clinically relevant cutoffs were chosen to guide clinical decision making. The final tool, Artificial Intelligence Prediction of Acute Leukemia (AI-PAL), was developed from this model. 1410 patients diagnosed with AML, APL, or ALL were included. Data quality control showed few missing values for each cohort, with the exception of uric acid and lactate dehydrogenase for the cohort from Hôpital Cochin. 679 patients from Hôpital Lyon Sud and Centre Hospitalier Universitaire de Clermont-Ferrand were split into the training (n=477) and internal testing (n=202) sets. 731 patients from the four other cohorts were used for external validation. Overall AUCs across all validation cohorts were 0·97 (95% CI 0·95-0·99) for APL, 0·90 (0·83-0·97) for ALL, and 0·89 (0·82-0·95) for AML. Cutoffs were then established on the overall cohort of 1410 patients to guide clinical decisions. Confident cutoffs showed two (0·14%) wrong predictions for ALL, four (0·28%) wrong predictions for APL, and three (0·21%) wrong predictions for AML. Use of the overall cutoff greatly reduced the number of missing predictions; diagnosis was proposed for 1375 (97·5%) of 1410 patients for each category, with only a slight increase in wrong predictions. The final model evaluation across both the internal testing and external validation sets showed accuracy of 99·5% for ALL diagnosis, 98·8% for AML diagnosis, and 99·7% for APL diagnosis in the confident model and accuracy of 87·9% for ALL diagnosis, 86·3% for AML diagnosis, and 96·1% for APL diagnosis in the overall model. AI-PAL allowed for accurate diagnosis of the three main acute leukaemia subtypes. Based on ten simple laboratory parameters, its broad availability could help guide initial therapies in a context where cytological expertise is lacking, such as in low-income countries. None.

Enhancing corn quality prediction: Variable selection and explainable AI in spectroscopic analysis

This study addresses the challenge of effectively selecting relevant variables and providing interpretable insights in spectroscopic analysis (1100–2498 nm) of corn quality (moisture, fat, protein, and starch), incorporating variable selection techniques and explainable artificial intelligence (AI). Three variable selection algorithms were used to select 36 important variables, and through combinatorial optimization, only 11 common wavelengths were chosen. Partial least squares regression (PLSR) models were developed using the full spectral range, individual variables, and common variables. Based on the root mean square errors of prediction (RMSEP), the PLSR models with common variables outperformed individual feature variables for starch (0.22 % vs. 0.24 %) and were comparable for oil (0.05 % vs. 0.05 %), water (0.02 % vs. 0.01 %) and protein (0.12 % vs. 0.09 %). The SHapley Additive exPlanation (SHAP) method was employed to explain the PLSR model and assess the contribution of common variables. This combined approach significantly improved the accuracy and interpretability of corn prediction models, providing trust in the results and facilitating the understanding of the relationship between spectral features and corn quality attributes.