Sequential Forward Selection Research Articles

A model based on machine learning for the prediction of cyclosporin A trough concentration in Chinese allo-HSCT patients

ABSTRACT Background Cyclosporin A is a calcineurin inhibitor which has a narrow therapeutic window and high interindividual variability. Various population pharmacokinetic models have been reported; however, professional software and technical personnel were needed and the variables of the models were limited. Therefore, the aim of this study was to establish a model based on machine learning to predict CsA trough concentrations in Chinese allo-HSCT patients. Methods A total of 7874 cases of CsA therapeutic drug monitoring data from 2069 allo-HSCT patients were retrospectively included. Sequential forward selection was used to select variable subsets, and eight different algorithms were applied to establish the prediction model. Results XGBoost exhibited the highest prediction ability. Except for the variables that were identified by previous studies, some rarely reported variables were found, such as norethindrone, WBC, PAB, and hCRP. The prediction accuracy within ±30% of the actual trough concentration was above 0.80, and the predictive ability of the models was demonstrated to be effective in external validation. Conclusion In this study, models based on machine learning technology were established to predict CsA levels 3–4 days in advance during the early inpatient phase after HSCT. A new perspective for CsA clinical application is provided.

Machine Learning-Based Classification of Subjective Cognitive Decline, Mild Cognitive Impairment, and Alzheimer's Dementia Using Neuroimage and Plasma Biomarkers.

Alzheimer's disease (AD) progresses relentlessly from the preclinical to the dementia stage. The process begins decades before the diagnosis of dementia. Therefore, it is crucial to detect early manifestations to prevent cognitive decline. Recent advances in artificial intelligence help tackle the complex high-dimensional data encountered in clinical decision-making. In total, we recruited 206 subjects, including 69 cognitively unimpaired, 40 subjective cognitive decline (SCD), 34 mild cognitive impairment (MCI), and 63 AD dementia (ADD). We included 3 demographic, 1 clinical, 18 brain-image, and 3 plasma biomarker (Aß1-42, Aß1-40, and tau protein) features. We employed the linear discriminant analysis method for feature extraction to make data more distinguishable after dimension reduction. The sequential forward selection method was used for feature selection to identify the 12 best features for machine learning classifiers. We used both random forest and support vector machine as classifiers. The area under the receiver operative curve (AUROC) was close to 0.9 between diseased (combining ADD and MCI) and the controls. AUROC was higher than 0.85 between SCD and controls, 0.90 between MCI and SCD, and above 0.85 between ADD and MCI. We can differentiate between adjacent phases of the AD spectrum with blood biomarkers and brain MR images with the help of machine learning algorithms.

Enhancing wind power forecast accuracy using the weather research and forecasting numerical model-based features and artificial neuronal networks

Forecasting with accuracy the quantity of energy produced by wind power plants is crucial to enabling its optimal integration into power systems and electricity markets. Despite the remarkable improvements in the wind forecasting systems in recent years, large errors can still be observed, especially for longer time horizons. This work focuses on identifying new numerical weather prediction (NWP)-based features aiming to improve the overall quality of wind power forecasts.The methodology also incorporates a sequential forward feature selection algorithm. This algorithm was designed to select iteratively the meteorological features which minimize the wind forecast errors.The methodology was applied separately to seven wind parks in Portugal with different climate characteristics. The proposed approach allowed a reduction between 13% and 37% in the root mean square errors of wind power forecasts, compared with a baseline scenario. While the meteorological features identified for each wind park showed similarities within regions with analogous wind power generation profiles, each wind park required specific meteorological parameters as input data to obtain the best performance. Thus, the results show to be crucial to select the most relevant features of a specific site to maximize the accuracy of a wind power forecast.

Load2Load: Day-ahead load forecasting at aggregated level

A reliable and accurate short-term load forecasting (STLF) helps utilities and energy providers deal with the challenges posed by supply and demand balance, higher penetration of renewable energies and the development of electricity markets with increasingly complex pricing strategies in future smart grids. Recent advances in deep learning have been successively utilized to STLF. However, there is no certain study that evaluates the performances of different STLF methods at an aggregated level on different datasets with different numbers of daily measurements.In this study, a deep learning STLF architecture called Load2Load is proposed for day-ahead forecasting. Different forecasting methods have been evaluated and compared on two datasets with different temporal resolutions and features. An additive ensemble method as well as a selective ensemble method that selects the outputs of different forecasters in an hourly manner are proposed. Moreover; a modified sequential forward feature selection algorithm is proposed, resulting in better performance with a much smaller number of features.Numerical results show that the proposed Load2Load architecture has a competing performance compared to other advanced forecasters. When used together with the proposed ensemble methods, Load2Load can significantly improve the forecasting performance. The proposed feature selection algorithm results in better performance for the majority of the cases while reducing the dimensionality. According to the results with two different datasets; the proposed methods are shown to be robust to temporal resolutions, feature types and sequence lengths.

Dependency maximization forward feature selection algorithms based on normalized cross-covariance operator and its approximated form for high-dimensional data

Supervised feature selection (FS) for classification aims at finding a more discriminative subset from original features, to facilitate classifier training, improve classification performance and enhance model interpretability. Primary normalized cross-covariance operator (NOCCO) is a nonlinear kernel-based dependency measure between features and labels, including two inverse matrices, and its approximated version (ANOCCO) is simplified via exploiting linear kernel for features, delta kernel for labels and Moore–Penrose inverse. In this paper, we apply NOCCO and ANOCCO to FS task. According to sequential forward selection, a forward NOCCO-based FS algorithm (i.e., FoNOCCO) is directly implemented via various accelerating strategies, but its computational complexity is extremely high. To this end, we propose its fast version via maximizing ANOCCO (i.e., FoANOCCO), where each candidate feature is evaluated efficiently according to column recursive algorithm for Moore–Penrose inverse. Theoretical analysis shows that evaluating a candidate feature needs a time complexity of O(N2) in FoANOCCO and O(N3) in FoNOCCO, where N is the number of instances. On eight small-size high-dimensional benchmark data sets, our detailed experiments illustrate that our two forward FS algorithms are superior to six state-of-the-art FS methods with four baseline classifiers, and FoANOCCO runs 70 times faster than FoNOCCO with a 1.79 % accuracy drop.

Logistic-ELM: a novel fault diagnosis method for rolling bearings

The fault diagnosis of rolling bearings is a critical technique to realize predictive maintenance for mechanical condition monitoring. In real industrial systems, the main challenges for the fault diagnosis of rolling bearings pertain to the accuracy and real-time requirements. Most existing methods focus on ensuring the accuracy, and the real-time requirement is often neglected. In this paper, considering both requirements, we propose a novel fast fault diagnosis method for rolling bearings, based on extreme learning machine (ELM) and logistic mapping, named logistic-ELM. First, we identify 14 kinds of time-domain features from the original vibration signals according to mechanical vibration principles and adopt the sequential forward selection strategy to select optimal features from them to ensure the basic predictive accuracy and efficiency. Next, we propose the logistic-ELM for fast fault classification, where the biases in ELM are omitted and the random input weights are replaced by the chaotic logistic mapping sequence which involves a higher uncorrelation to obtain more accurate results with fewer hidden neurons. We conduct extensive experiments on the rolling bearing vibration signal dataset of the Case Western Reserve University bearing data centre. The experimental results show that the proposed approach outperforms existing state-of-the-art comparison methods in terms of the predictive accuracy, and the highest accuracies are 100%, 99.71%, 98%, 100%, 100%, and 100%, respectively, in seven separate sub data environments. Moreover, in terms of the runtime cost, the experimental results indicate that the proposed logistic-ELM can predict the fault in 40 ms with a high accuracy, up to 21-1858 times more rapid than existing methods based on support vector machine, convolutional neural network and multi-scale entropy. Other experiments of fault diagnosis of the rolling bearings under four different loads also indicate that the logistic-ELM can adapt to different operation conditions with high efficiency. The relevant code is publicly available at https://github.com/TAN-OpenLab/logistic-ELM .

Can CBCT-Based Delta Radiomics Predict Normal Lung Toxicity during Thoracic Radiation?

<h3>Purpose/Objective(s)</h3> Delta radiomics which refers to longitudinal changes of radiomic features over time has shown the potential to predict treatment response. However, its role in predicting normal lung toxicity has not been studied extensively. This study evaluates the potential for CBCT based delta radiomics in predicting radiotherapy induced lung parenchymal changes during thoracic radiotherapy. <h3>Materials/Methods</h3> Sixty-two patients were selected who underwent thoracic radiation with volume-modulated-arc-therapy. CBCT images from treatment weeks 1, 3 and 6 (CBCTfirst, CBCTmid, CBCTend) were included. An experienced radiologist reviewed the serial lung changes and geographic location. 19 patients had radiological confirmed early pneumonitis. The normal lung was delineated into regions receiving dose between 5-10 Gy(V5-10), 10-20 Gy(V10-20), above 20 Gy (V20) on the planning CT and structures were transferred to CBCT following rigid registration using the treatment planning system. Each CBCT was processed with resampling (2mm3 isotropic), feature extraction (103 Pyradiomic features) and normalization (z-score). Optimal delta features were selected following a forward sequential selection method. Dataset was resampled with synthetic minority oversampling technique. Support vector machine classifier was used to build a machine-learning-based model to pick up incidence of early pneumonitis. The models used were built using 80:20 training/testing data split and performance was evaluated by the area-under-the-receiver-operating-characteristic-curve (AUC). <h3>Results</h3> The median target dose received by the patients during thoracic radiation was 54 Gy. The delta features selected for model training following the sequential feature selection technique were SurfaceArea, Contrast, RootMeanSquared, MeanAbsoluteDeviation, SmallAreaHighGrayLevelEmphasis (GLZSM). At CBCTmid, support vector machine classifier resulted in the best performance in predicting early pneumonitis in the region V5-10, AUC being 0.89 (95% CI 0.89-0.903) and 0.76 (0.74-0.78) in the training and testing datasets, respectively. This was followed by V20 region in the CBCTend with AUC of 0.904(0.89-0.92) and 0.71(0.69-0.74) in the training and testing datasets, respectively. <h3>Conclusion</h3> This study supports a potential role of delta radiomics to predict incidence of early pneumonitis changes with reasonable accuracy in patients undergoing thoracic radiation. Future studies with more patients and external validation are needed to confirm these results and for clinical implementation. In future, prediction models for clinical pneumonitis using CBCT based delta radiomics will be explored.

An individualized medication model of sodium valproate for patients with bipolar disorder based on machine learning and deep learning techniques.

Valproic acid/sodium valproate (VPA) is a widely used anticonvulsant drug for maintenance treatment of bipolar disorders. In order to balance the efficacy and adverse events of VPA treatment, an individualized dose regimen is necessary. This study aimed to establish an individualized medication model of VPA for patients with bipolar disorder based on machine learning and deep learning techniques. The sequential forward selection (SFS) algorithm was applied for selecting a feature subset, and random forest was used for interpolating missing values. Then, we compared nine models using XGBoost, LightGBM, CatBoost, random forest, GBDT, SVM, logistic regression, ANN, and TabNet, and CatBoost was chosen to establish the individualized medication model with the best performance (accuracy = 0.85, AUC = 0.91, sensitivity = 0.85, and specificity = 0.83). Three important variables that correlated with VPA daily dose included VPA TDM value, antipsychotics, and indirect bilirubin. SHapley Additive exPlanations was applied to visually interpret their impacts on VPA daily dose. Last, the confusion matrix presented that predicting a daily dose of 0.5g VPA had a precision of 55.56% and recall rate of 83.33%, and predicting a daily dose of 1g VPA had a precision of 95.83% and a recall rate of 85.19%. In conclusion, the individualized medication model of VPA for patients with bipolar disorder based on CatBoost had a good prediction ability, which provides guidance for clinicians to propose the optimal medication regimen.

Applying Machine Learning to Predict Esophageal Cancer Recurrence after Esophagectomy

INTRODUCTION: Artificial intelligence and machine learning (ML) models have recently been adapted in healthcare applications with promising results. The objective of this proof-of-concept study was to develop a ML model designed to predict esophageal cancer recurrence after esophagectomy. METHODS: We conducted a retrospective study of 260 consecutive patients who underwent esophagectomy for esophageal cancer from 2009 through 2018. Over 20,000 patient-specific characteristics were collected. Risk prediction models for different prediction windows were constructed via a sequential forward selection process. Five traditional machine learning algorithms including Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), and Naive Bayes (NB) were included in our analysis. Model performance was assessed by calculating sensitivity, specificity, positive predictive value (PPV), F1 score, area under the receiver operating characteristic curve (AUC), and overall accuracy using five-fold cross-validation. RESULTS: Of the 260 patients, 121 (46.5%) experienced cancer recurrence at a median of 267 days (49-2,057 days). The highest AUC for prediction of recurrence at any time after esophagectomy was 0.82 (SD ± 0.02) by DT. Similarly, the accuracy of the model to predict whether or not a new subject inserted into the model will have recurrence at any time was 82% (0.82 ± 0.03) by RF. The LR algorithm had the highest accuracy (92%, 0.92 ± 0.01) for predicting recurrence within 180 days after esophagectomy (Table). CONCLUSION: This proof-of-concept study demonstrates the feasibility of using machine learning models to predict esophageal cancer recurrence. The accuracy and AUC of the machine learning models exceeded 80% in all recurrence prediction timeframes. Table. - Overall Performance of the Machine Learning Models Prediction window No. (%) of subjects with recurrence/no. (%) of subjects without recurrence Model Accuracy (SD) AUC (SD) Recurrence at any time after esophagectomy 121 (46.5) / 139 (53.5) LR 0.80 (0.05) 0.81 (0.04) SVM 0.77 (0.05) 0.79 (0.06) RF 0.82 (0.03) 0.82 (0.05) DT 0.79 (0.03) 0.82 (0.02) NB 0.80 (0.04) 0.78 (0.06) Recurrence within 180 days after esophagectomy 37 (14.2) / 223 (85.8) LR 0.92 (0.01) 0.78 (0.05) SVM 0.92 (0.02) 0.81 (0.07) RF 0.90 (0.03) 0.73 (0.10) DT 0.90 (0.02) 0.70 (0.04)

Integrating machine learning with electronic health record data to facilitate detection of prolactin level and pharmacovigilance signals in olanzapine-treated patients.

Background and aimAvailable evidence suggests elevated serum prolactin (PRL) levels in olanzapine (OLZ)-treated patients with schizophrenia. However, machine learning (ML)-based comprehensive evaluations of the influence of pathophysiological and pharmacological factors on PRL levels in OLZ-treated patients are rare. We aimed to forecast the PRL level in OLZ-treated patients and mine pharmacovigilance information on PRL-related adverse events by integrating ML and electronic health record (EHR) data.MethodsData were extracted from an EHR system to construct an ML dataset in 672×384 matrix format after preprocessing, which was subsequently randomly divided into a derivation cohort for model development and a validation cohort for model validation (8:2). The eXtreme gradient boosting (XGBoost) algorithm was used to build the ML models, the importance of the features and predictive behaviors of which were illustrated by SHapley Additive exPlanations (SHAP)-based analyses. The sequential forward feature selection approach was used to generate the optimal feature subset. The co-administered drugs that might have influenced PRL levels during OLZ treatment as identified by SHAP analyses were then compared with evidence from disproportionality analyses by using OpenVigil FDA.ResultsThe 15 features that made the greatest contributions, as ranked by the mean (|SHAP value|), were identified as the optimal feature subset. The features were gender_male, co-administration of risperidone, age, co-administration of aripiprazole, concentration of aripiprazole, concentration of OLZ, progesterone, co-administration of sulpiride, creatine kinase, serum sodium, serum phosphorus, testosterone, platelet distribution width, α-L-fucosidase, and lipoprotein (a). The XGBoost model after feature selection delivered good performance on the validation cohort with a mean absolute error of 0.046, mean squared error of 0.0036, root-mean-squared error of 0.060, and mean relative error of 11%. Risperidone and aripiprazole exhibited the strongest associations with hyperprolactinemia and decreased blood PRL according to the disproportionality analyses, and both were identified as co-administered drugs that influenced PRL levels during OLZ treatment by SHAP analyses.ConclusionsMultiple pathophysiological and pharmacological confounders influence PRL levels associated with effective treatment and PRL-related side-effects in OLZ-treated patients. Our study highlights the feasibility of integration of ML and EHR data to facilitate the detection of PRL levels and pharmacovigilance signals in OLZ-treated patients.

A New Feature Analysis Approach to Selecting Channels of EEG for Fatigue Driving.

Fatigued driving is a significant contributor to traffic accidents. There are some issues with common EEG data of 32 channels, 64 channels, and 128 channels, such as difficult acquisition, high data redundancy, and difficult practical application. A new channel selection method called ReliefF_SFS is proposed to address the problem of how to reduce the number of channels while maintaining classification accuracy. It combines the ReliefF algorithm and the sequential forward selection (SFS) algorithm. When only T6, O1, Oz, T4, P3, and FC3 are used, the classification accuracy under Theta_Std+FE combined with ReliefF_SFS achieves 99.45%. The strategy suggested in this paper not only ensures the recognition accuracy but also reduces the number of channels when compared to other models based on the same data set.

Cardiac magnetic resonance radiomics for prediction of incident heart failure: a feasibility study in the UK Biobank Imaging cohort

Abstract Background Cardiac magnetic resonance (CMR) radiomics is a novel image quantification technique with the potential to improve image-based disease diagnosis and prediction. Purpose In this proof-of-concept study, we aimed to evaluate the utility of CMR radiomics in the prediction of incident heart failure (HF). Methods We studied 32,121 UK Biobank participants with CMR. Incident HF was defined from linked Hospital Episode Statistics. To create a balanced cohort, we identified as comparators an equal number of randomly selected subjects who did not develop the outcome of interest during this period. Radiomics shape, first-order and texture features were extracted from short-axis cine images (left and right ventricle, left ventricular myocardium) using the Pyradiomics toolbox. Vascular risk factors (VRFs) were considered as additional predictors. Feature selection was conducted using the sequential forward selection technique and modelling was performed using Support Vector Machine (SVM) methods with 5-fold cross-validation. Models were developed using 1) VRFs alone, 2) radiomics alone, and 3) VRFs and radiomics. We determined model performance using receiver operating characteristic (ROC) curve and area under the curve (AUC) scores. Results Over average follow-up time of 3.7 (±1.3) years, 209 participants experienced incident HF. Among vascular risk factors, age, body size, hypertension, diabetes, high cholesterol were chosen for the incident HF predictive model (Accuracy: 0.66, AUC: 0.73) by the SVM methods. The model based on radiomics features reached a marginal improvement compared to vascular risk factors alone (Accuracy: 0.71, AUC: 0.75). The combination of VRFs and radiomics features significantly improved the performance of the model to predict incident HF compared to VRFs alone (Accuracy: 0.77; AUC: 0.83; p&amp;lt;0.05) Conclusion We demonstrate the feasibility of CMR radomics features to predict incident HF and illustrate their added value over vascular risk factors. Funding Acknowledgement Type of funding sources: Public grant(s) – EU funding.

Computational prediction of Calu-3-based in vitro pulmonary permeability of chemicals

Pulmonary is a potential route for drug delivery and exposure to toxic chemicals. The human bronchial epithelial cell line Calu-3 is generally considered to be a useful in vitro model of pulmonary permeability by calculating the apparent permeability coefficient (Papp) values. Since in vitro experiments are time-consuming and labor-intensive, computational models for pulmonary permeability are desirable for accelerating drug design and toxic chemical assessment. This study presents the first attempt for developing quantitative structure-activity relationship (QSAR) models for addressing this goal. A total of 57 chemicals with Papp values based on Calu-3 experiments was first curated from literature for model development and testing. Subsequently, eleven descriptors were identified by a sequential forward feature selection algorithm to maximize the cross-validation performance of a voting regression model integrating linear regression and nonlinear random forest algorithms. With applicability domain adjustment, the developed model achieved high performance with correlation coefficient values of 0.935 and 0.824 for cross-validation and independent test, respectively. The preliminary results showed that computational models could be helpful for predicting Calu-3-based in vitro pulmonary permeability of chemicals. Future works include the collection of more data for further validating and improving the model.

A Data-Driven Short-Term PV Generation and Load Forecasting Approach for Microgrid Applications

The data-driven (DD) is a systematic approach to improve the data and model by deriving/adding features to address the problem identified during the iterative loop of forecasting model development. This article proposes a DD framework for forecasting short-term PV generation and load demand. A framework of three stages with a unique contribution in each stage, such as generalizing data preprocessing steps (stage-1), multivariate feature generation and selection (stage-2), and model hyperparameter tuning (stage-3) for further improvement in forecasting is proposed. It focuses on data as well as forecasting models. The whole process is analyzed using the time-series measured data collected from a real-life demonstration project in Ireland. Data preprocessing is generalized for both generation and demand forecasting under the same framework. The relevant features are selected with the help of the proposed random forest sequential forward feature selection algorithm. Hyperparameters are tuned through tree-structured Parzen estimator algorithm for further improvement. In addition, the performance of the classical autoregressive integrated moving average model is compared with the machine learning-based gate recurrent unit, long short term memory, recurrent neural network, and convolutional neural network models. Results show that the data-driven forecasting model framework systematically improves the model performance. The seasonal variation has also a high impact on the model performances.

Clustering-based fusion for medical information retrieval

Medicine is a fast-moving field, and the number of medical publications has increased rapidly over recent years. How to find relevant information from this vast pool of research effectively and efficiently has therefore become highly challenges. Previous studies have demonstrated that data fusion can improve search performance if properly utilized. However, in most cases effectiveness is the only concern and efficiency is not considered. A fusion-based system is by nature more complicated and expensive computationally than other retrieval models such as BM25, because many component retrieval systems and an extra layer of fusion are required. The number of component retrieval systems involved is an important indicator of complexity of the fusion-based system. We aim to select the optimal k-subset of component retrieval systems for any given number k, to optimize both fusion performance and reduce the cost of data fusion. A clustering-based approach is proposed. First all the candidates are divided into clusters by the Chameleon clustering algorithm, then representatives from every cluster are chosen by Sequential Forward Selection for fusion. Evaluated with two datasets from TREC, the proposed method performs more effectively than the other baseline methods including the state-of-the-art subset selection method significantly. When either of the two typical fusion methods is used, an improvement rate of over 10% is observed for both measures Mean Average Precision and Recall-level Precision, and an improvement rate of over 5% is observed for both measures Precision at 10 document level and Mean Reciprocal Rank.

Using Machine Learning for Nutrient Content Detection of Aquaponics-Grown Plants Based on Spectral Data

Nutrients derived from fish feed are insufficient for optimal plant growth in aquaponics; therefore, they need to be supplemented. Thus, estimating the amount of supplementation needed can be achieved by looking at the nutrient contents of the plant. This study aims to develop trustworthy machine learning models to estimate the nitrogen (N), phosphorus (P), and potassium (K) contents of aquaponically grown lettuce. A FieldSpec4, Pro FR portable spectroradiometer (ASD Inc., Analytical Spectral Devices Boulder, Boulder, CO, USA) was used to measure leaf reflectance spectra, and 128 lettuce seedlings given four NPK treatments were used for spectra acquisition and total NPK estimation. Principal component analysis (PCA), genetic algorithms (GA), and sequential forward selection (SFS) were applied to select the optimal wavebands. Partial least squares regression (PLSR), back-propagation neural network (BPNN), and random forest (RF) approaches were used to develop the predictive models of NPK contents using the selected optimal wavelengths. Good and significantly correlated predictive accuracy was obtained in comparison with the laboratory-measured freshly cut lettuce leaves with R2 ≥ 0.94. The proposed approach provides a pathway toward automatic nutrient estimation of aquaponically grown lettuce. Consequently, aquaponics will become more intelligent, and will be adopted as a precision agriculture technology.

Detection and Classification of Unannounced Physical Activities and Acute Psychological Stress Events for Interventions in Diabetes Treatment

Detection and classification of acute psychological stress (APS) and physical activity (PA) in daily lives of people with chronic diseases can provide precision medicine for the treatment of chronic conditions such as diabetes. This study investigates the classification of different types of APS and PA, along with their concurrent occurrences, using the same subset of feature maps via physiological variables measured by a wristband device. Random convolutional kernel transformation is used to extract a large number of feature maps from the biosignals measured by a wristband device (blood volume pulse, galvanic skin response, skin temperature, and 3D accelerometer signals). Three different feature selection techniques (principal component analysis, partial least squares–discriminant analysis (PLS-DA), and sequential forward selection) as well as four approaches for addressing imbalanced sizes of classes (upsampling, downsampling, adaptive synthetic sampling (ADASYN), and weighted training) are evaluated for maximizing detection and classification accuracy. A long short-term memory recurrent neural network model is trained to estimate PA (sedentary state, treadmill run, stationary bike) and APS (non-stress, emotional anxiety stress, mental stress) from wristband signals. The balanced accuracy scores for various combinations of data balancing and feature selection techniques range between 96.82% and 99.99%. The combination of PLS–DA for feature selection and ADASYN for data balancing provide the best overall performance. The detection and classification of APS and PA types along with their concurrent occurrences can provide precision medicine approaches for the treatment of diabetes.

An interpretable stacking ensemble learning framework based on multi-dimensional data for real-time prediction of drug concentration: The example of olanzapine.

Background and Aim: Therapeutic drug monitoring (TDM) has evolved over the years as an important tool for personalized medicine. Nevertheless, some limitations are associated with traditional TDM. Emerging data-driven model forecasting [e.g., through machine learning (ML)-based approaches] has been used for individualized therapy. This study proposes an interpretable stacking-based ML framework to predict concentrations in real time after olanzapine (OLZ) treatment. Methods: The TDM-OLZ dataset, consisting of 2,142 OLZ measurements and 472 features, was formed by collecting electronic health records during the TDM of 927 patients who had received OLZ treatment. We compared the performance of ML algorithms by using 10-fold cross-validation and the mean absolute error (MAE). The optimal subset of features was analyzed by a random forest-based sequential forward feature selection method in the context of the top five heterogeneous regressors as base models to develop a stacked ensemble regressor, which was then optimized via the grid search method. Its predictions were explained by using local interpretable model-agnostic explanations (LIME) and partial dependence plots (PDPs). Results: A state-of-the-art stacking ensemble learning framework that integrates optimized extra trees, XGBoost, random forest, bagging, and gradient-boosting regressors was developed for nine selected features [i.e., daily dose (OLZ), gender_male, age, valproic acid_yes, ALT, K, BW, MONO#, and time of blood sampling after first administration]. It outperformed other base regressors that were considered, with an MAE of 0.064, R-square value of 0.5355, mean squared error of 0.0089, mean relative error of 13%, and ideal rate (the percentages of predicted TDM within ± 30% of actual TDM) of 63.40%. Predictions at the individual level were illustrated by LIME plots, whereas the global interpretation of associations between features and outcomes was illustrated by PDPs. Conclusion: This study highlights the feasibility of the real-time estimation of drug concentrations by using stacking-based ML strategies without losing interpretability, thus facilitating model-informed precision dosing.

Feature selection using Decomposed Mutual Information Maximization

Feature selection has been recognized for long as an important preprocessing technique to reduce dimensionality and improve the performance of regression and classification tasks. The class of sequential forward feature selection methods based on Mutual Information (MI) is widely used in practice, mainly due to its computational efficiency and independence from the specific classifier. A recent work introduced a theoretical framework for this class of methods which explains the existing proposals as approximations to an optimal target objective function. Such framework made clear the advantages and drawbacks of each proposal. Methods accounting for the redundancy of candidate features using a maximization function and considering the so-called complementary effect are among the best ones. However, they still penalize the complementarity, which is an important drawback. This paper proposes the Decomposed Mutual Information Maximization (DMIM) method, which keeps the good theoretical properties of the best methods proposed so far but overcomes the complementarity penalization by applying the maximization separately to the inter-feature and class-relevant redundancies. DMIM was extensively evaluated and compared with other methods, both theoretically and using two synthetic scenarios and 20 publicly available real datasets applied to specific classifiers. Our results show that DMIM achieves a better classification performance than the remaining forward feature selection methods based on MI.

A Diagnostic Model of Breast Cancer Based on Digital Mammogram Images Using Machine Learning Techniques

Breast cancer disease is one of the most recorded cancers that lead to morbidity and maybe death among women around the world. Recent research statistics have exposed that one from 8 females in the USA and one from 10 females in Europe are contaminated by breast cancer. The challenge with this disease is how to develop a relaxed and fast diagnosing method. One of the attractive ways of early breast cancer diagnosis is based on the mammogram images analysis of the breast using a computer-aided diagnosing (CAD) tool. This paper firstly aimed to propose an efficient method for diagnosing tumors based on mammogram images of breasts using a machine learning approach. Secondly, this paper aimed to the development of a CAD software program for breast cancer diagnosing based on the proposed method in the first step. The followed step-by-step procedure of the proposed method is performed by passing the Mammographic Image Analysis Society (MIAS) through five steps of image preprocessing, image segmentation using seeded region growing (SRG) algorithm, feature extraction using different feature’s extraction classes, and important and effectiveness feature selection using the Sequential Forward Selection (SFS) technique, and finally, the Support Vector Machine (SVM) algorithm is used as a binary classifier in two classification levels. The first level classifier is used to categorize the given image as normal or abnormal while the second-level classifier is used for further classifying the abnormal image as either a malignant or benign cancer. The proposed method is studied and investigated in two phases: the training phase and the testing phase, with the MIAS dataset of mammogram images, using 70% and 30% ratios of dataset images for the training and testing sets, respectively. The practical implementation of the proposed method and the graphical user interface (GUI) CAD tool are carried out using MATLAB software. Experimental results of the proposed method have shown that the accuracy of the proposed method reached 100% in classifying images as normal and abnormal mammogram images while the classification accuracy for benign and malignant is equal to 87.1%.