Interpretable Machine Learning Research Articles

Interpretable Machine Learning: A Case Study on Predicting Fuel Consumption in VLGC Ship Propulsion

The integration of machine learning (ML) in marine engineering has been increasingly subjected to stringent regulatory scrutiny. While environmental regulations aim to reduce harmful emissions and energy consumption, there is also a growing demand for the interpretability of ML models to ensure their reliability and adherence to safety standards. This research highlights the need to develop models that are both transparent and comprehensible to domain experts and regulatory bodies. This paper underscores the importance of transparency in machine learning through a use case involving a VLGC ship two-stroke propulsion engine. By adhering to the CRISP-DM standard, we fostered close collaboration between marine engineers and machine learning experts to circumvent the common pitfalls of automated ML. The methodology included comprehensive data exploration, cleaning, and verification, followed by feature selection and training of linear regression and decision tree models that are not only transparent but also highly interpretable. The linear model achieved an RMSE of 23.16 and an MRAE of 14.7%, while the accuracy of decision trees ranged between 96.4% and 97.69%. This study demonstrates that machine learning models for predicting propulsion engine fuel consumption can be interpretable, adhering to regulatory requirements, while still achieving adequate predictive performance.

Interpretable machine‐learning models for predicting creep recovery of concrete

AbstractCreep recovery of concrete is essential for accurately assessing the performance of concrete structures over service time. Existing creep recovery models exhibit low accuracy, and the influencing factors of creep recovery remain inadequately elucidated. In this paper, interpretable machine learning (ML) techniques were employed to develop a prediction model for concrete creep recovery. Several ML techniques were selected including random forest (RF), support vector regression (SVR), extreme gradient boosting (XGBoost) and light gradient boosting machine (LGBM). In order to maximize the sample size of the dataset, 109 sets of creep recovery data were collected from existing literatures for model training. Feature selection is utilized to determine the input parameters for ML models, and 12 input variables were selected. The model is fine‐tuned using Bayesian optimization techniques. To ensure the reliability of ML models, 10‐fold cross‐validation and random data splitting were implemented. The results indicate that the ML models exhibited higher accuracy compared to the existing creep recovery model. Among these ML models, LGBM demonstrated superior accuracy, efficiency and stability (with R2 = 0.993, 0.978, and 0.973 for the training, testing, and validation sets, respectively). Shapley additive explanations (SHAP) were employed to interpret the significance of each input parameter on ML model prediction. Duration after unloading, stress magnitude, and ambient relative humidity were the main feature variables influencing concrete creep recovery. Upon comparing the influencing factors, it was discerned that there exists a distinct difference between creep and creep recovery of concrete.

Developing a Wearable Sensor-Based Digital Biomarker of Opioid Dependence.

Repeated opioid exposure leads to a variety of physiologic adaptations that develop at different rates and may foreshadow risk of opioid-use disorder (OUD), including dependence and withdrawal. Digital pharmacovigilance strategies that use noninvasive sensors to identify physiologic adaptations to opioid use represent a novel strategy to facilitate safer opioid prescribing. This study aims to identify wearable sensor-derived features associated with opioid dependence by comparing opioid-naïve individuals to chronic opioid users with acute pain and developing a machine-learning model to distinguish between the 2 groups. Using a longitudinal observational study design, continuous physiologic data were collected on participants with acute pain receiving opioid analgesia. Monitoring continued throughout hospitalization and for up to 7 days posthospital discharge. Opioid administration data were obtained from electronic health record (EHR) and participant self-report. Participants were classified as belonging to 1 of 3 categories based on opioid use history: naïve, occasional, or chronic use. Thirty features were derived from sensor data, and an additional 9 features were derived from participant demographic and treatment characteristics. Physiologic feature behavior immediately postopioid use was compared among naïve and chronic participants, and subsequently features were used to generate machine learning models which were validated using cross-validation and holdout data. Forty-one participants with a combined total of 169 opioid administrations were ultimately included in the final analysis. Four interpretable decision tree-based machine learning models with 14 sensor-based and 5 clinical features were developed to predict class membership on the level of a given observation (dose) and on the participant level. Ranges for model metrics on the participant level were as follows: accuracy 70% to 90%, sensitivity 67% to 100%, and specificity 67% to 100%. Wearable sensor-derived digital biomarkers can be used to predict opioid use status (naïve versus chronic) and the differentiating features may be detecting opioid dependence. Future work should be aimed at further delineating the phenomenon identified in these models (including opioid dependence and/or withdrawal) and at identifying transition states where an individual changes from 1 profile to another with repetitive opioid exposure.

Using interpretable machine learning to predict bloodstream infection and antimicrobial resistance in patients admitted to ICU: Early alert predictors based on EHR data to guide antimicrobial stewardship.

Nosocomial infections and Antimicrobial Resistance (AMR) stand as formidable healthcare challenges on a global scale. To address these issues, various infection control protocols and personalized treatment strategies, guided by laboratory tests, aim to detect bloodstream infections (BSI) and assess the potential for AMR. In this study, we introduce a machine learning (ML) approach based on Multi-Objective Symbolic Regression (MOSR), an evolutionary approach to create ML models in the form of readable mathematical equations in a multi-objective way to overcome the limitation of standard single-objective approaches. This method leverages readily available clinical data collected upon admission to intensive care units, with the goal of predicting the presence of BSI and AMR. We further assess its performance by comparing it to established ML algorithms using both naturally imbalanced real-world data and data that has been balanced through oversampling techniques. Our findings reveal that traditional ML models exhibit subpar performance across all training scenarios. In contrast, MOSR, specifically configured to minimize false negatives by optimizing also for the F1-Score, outperforms other ML algorithms and consistently delivers reliable results, irrespective of the training set balance with F1-Score.22 and.28 higher than any other alternative. This research signifies a promising path forward in enhancing Antimicrobial Stewardship (AMS) strategies. Notably, the MOSR approach can be readily implemented on a large scale, offering a new ML tool to find solutions to these critical healthcare issues affected by limited data availability.

Tree smoothing: Post-hoc regularization of tree ensembles for interpretable machine learning

Random Forests (RFs) are powerful ensemble learning algorithms that are widely used in various machine learning tasks. However, they tend to overfit noisy or irrelevant features, which can result in decreased generalization performance. Post-hoc regularization techniques aim to solve this problem by modifying the structure of the learned ensemble after training. We propose a novel post-hoc regularization via tree smoothing for classification tasks to leverage the reliable class distributions closer to the root node whilst reducing the impact of more specific and potentially noisy splits deeper in the tree. Our novel approach allows for a form of pruning that does not alter the general structure of the trees, adjusting the influence of nodes based on their proximity to the root node. We evaluated the performance of our method on various machine learning benchmark data sets and on cancer data from The Cancer Genome Atlas (TCGA). Our approach demonstrates competitive performance compared to the state-of-the-art and, in the majority of cases, and outperforms it in most cases in terms of prediction accuracy, generalization, and interpretability.

Interpretable Clinical Decision-Making Application for Etiological Diagnosis of Ventricular Tachycardia Based on Machine Learning.

Background: Ventricular tachycardia (VT) can broadly be categorised into ischemic heart disease, non-ischemic structural heart disease, and idiopathic VT. There are few studies related to the application of machine learning for the etiological diagnosis of VT, and the interpretable methods are still in the exploratory stage for clinical decision-making applications. Objectives: The aim is to propose a machine learning model for the etiological diagnosis of VT. Interpretable results based on models are compared with expert knowledge, and interpretable evaluation protocols for clinical decision-making applications are developed. Methods: A total of 1305 VT patient data from 1 January 2013 to 1 September 2023 at the Arrhythmia Centre of Fuwai Hospital were included in the study. Clinical data collected during hospitalisation included demographics, medical history, vital signs, echocardiographic results, and laboratory test outcomes. Results: The XGBoost model demonstrated the best performance in VT etiological diagnosis (precision, recall, and F1 were 88.4%, 88.5%, and 88.4%, respectively). A total of four interpretable machine learning methods applicable to clinical decision-making were evaluated in terms of visualisation, clinical usability, clinical applicability, and efficiency with expert knowledge interpretation. Conclusions: The XGBoost model demonstrated superior performance in the etiological diagnosis of VT, and SHAP and decision tree interpretable methods are more favoured by clinicians for decision-making.

Exploring the correlation between DNA methylation and biological age using an interpretable machine learning framework

DNA methylation plays a significant role in regulating transcription and exhibits a systematic change with age. These changes can be used to predict an individual’s age. First, to identify methylation sites associated with biological age; second, to construct a biological age prediction model and preliminarily explore the biological significance of methylation-associated genes using machine learning. A biological age prediction model was constructed using human methylation data through data preprocessing, feature selection procedures, statistical analysis, and machine learning techniques. Subsequently, 15 methylation data sets were subjected to in-depth analysis using SHAP, GO enrichment, and KEGG analysis. XGBoost, LightGBM, and CatBoost identified 15 groups of methylation sites associated with biological age. The cg23995914 locus was identified as the most significant contributor to predicting biological age by calculating SHAP values. Furthermore, GO enrichment and KEGG analyses were employed to initially explore the methylated loci’s biological significance.

StratoMod: predicting sequencing and variant calling errors with interpretable machine learning

Despite the variety in sequencing platforms, mappers, and variant callers, no single pipeline is optimal across the entire human genome. Therefore, developers, clinicians, and researchers need to make tradeoffs when designing pipelines for their application. Currently, assessing such tradeoffs relies on intuition about how a certain pipeline will perform in a given genomic context. We present StratoMod, which addresses this problem using an interpretable machine-learning classifier to predict germline variant calling errors in a data-driven manner. We show StratoMod can precisely predict recall using Hifi or Illumina and leverage StratoMod’s interpretability to measure contributions from difficult-to-map and homopolymer regions for each respective outcome. Furthermore, we use Statomod to assess the effect of mismapping on predicted recall using linear vs. graph-based references, and identify the hard-to-map regions where graph-based methods excelled and by how much. For these we utilize our draft benchmark based on the Q100 HG002 assembly, which contains previously-inaccessible difficult regions. Furthermore, StratoMod presents a new method of predicting clinically relevant variants likely to be missed, which is an improvement over current pipelines which only filter variants likely to be false. We anticipate this being useful for performing precise risk-reward analyses when designing variant calling pipelines.

Machine learning interpretability methods to characterize the importance of hematologic biomarkers in prognosticating patients with suspected infection

ObjectiveTo evaluate the effectiveness of Monocyte Distribution Width (MDW) in predicting sepsis outcomes in emergency department (ED) patients compared to other hematologic parameters and vital signs, and to determine whether routine parameters could substitute MDW in machine learning models. MethodsWe conducted a retrospective analysis of data from 10,229 ED patients admitted to a large regional safety-net hospital in Cleveland, Ohio who had suspected infections and developed sepsis-associated poor outcomes. We developed a new analytical framework consisting of seven data models and an ensemble of high accuracy machine learning (ML) algorithms (accuracy values ranging from 0.83 to 0.90) to predict sepsis-associated poor outcomes (3-day intensive care unit stay or death). Local Interpretable Model-Agnostic Explanation (LIME) and Shapley Additive Value (SHAP) interpretability methods were utilized to assess the contributions of individual hematologic parameters. ResultsThe ML interpretability analysis indicated that the predictive value of MDW is significantly reduced when other hematological parameters and vital signs are considered. The results suggest that complete blood count with differential (CBD-DIFF) alongside vital signs can effectively replace MDW in high accuracy machine learning algorithms for screening poor outcome associated with sepsis. ConclusionMDW, although a newly approved biomarker for sepsis, does not significantly enhance prediction models when combined with routinely available parameters and vital signs. Hospitals, especially those with resource constraints, can rely on existing parameters with high accuracy machine learning models to predict sepsis outcomes effectively, thereby reducing the need for specialized tests like MDW.

A Knowledge–Data Dual‐Driven Framework for Predicting the Molecular Properties of Rechargeable Battery Electrolytes

Developing rechargeable batteries that operate within a wide temperature range and possess high safety has become necessary with increasing demands. Rapid and accurate assessment of the melting points (MPs), boiling points (BPs), and flash points (FPs) of electrolyte molecules is essential for expediting battery development. Herein, we introduce Knowledge‐based electrolyte Property prediction Integration (KPI), a knowledge–data dual‐driven framework for molecular property prediction of electrolytes. Initially, the KPI collects molecular structures and properties, and then automatically organizes them into structured datasets. Subsequently, interpretable machine learning further explores the structure–property relationships of molecules from a microscopic perspective. Finally, by embedding the discovered knowledge into property prediction models, the KPI achieved very low mean absolute errors of 10.4, 4.6, and 4.8 K for MP, BP, and FP predictions, respectively. The KPI reached state‐of‐the‐art results in 18 out of 20 datasets. Utilizing molecular neighbor search and high‐throughput screening, 15 and 14 promising molecules, with and without Chemical Abstracts Service Registry Number, respectively, were predicted for wide‐temperature‐range and high‐safety batteries. The KPI not only accurately predicts molecular properties and deepens the understanding of structure–property relationships but also serves as an efficient framework for integrating artificial intelligence and domain knowledge.

A Knowledge-Data Dual-Driven Framework for Predicting the Molecular Properties of Rechargeable Battery Electrolytes.

Developing rechargeable batteries that operate within a wide temperature range and possess high safety has become necessary with increasing demands. Rapid and accurate assessment of the melting points (MPs), boiling points (BPs), and flash points (FPs) of electrolyte molecules is essential for expediting battery development. Herein, we introduce Knowledge-based electrolyte Property prediction Integration (KPI), a knowledge-data dual-driven framework for molecular property prediction of electrolytes. Initially, the KPI collects molecular structures and properties, and then automatically organizes them into structured datasets. Subsequently, interpretable machine learning further explores the structure-property relationships of molecules from a microscopic perspective. Finally, by embedding the discovered knowledge into property prediction models, the KPI achieved very low mean absolute errors of 10.4, 4.6, and 4.8 K for MP, BP, and FP predictions, respectively. The KPI reached state-of-the-art results in 18 out of 20 datasets. Utilizing molecular neighbor search and high-throughput screening, 15 and 14 promising molecules, with and without Chemical Abstracts Service Registry Number, respectively, were predicted for wide-temperature-range and high-safety batteries. The KPI not only accurately predicts molecular properties and deepens the understanding of structure-property relationships but also serves as an efficient framework for integrating artificial intelligence and domain knowledge.

Integrated analysis of gene expressions and targeted mirnas for explaining crosstalk between oral and esophageal squamous cell carcinomas through an interpretable machine learning approach.

This study explores the bidirectional relation of esophageal squamous cell carcinoma (ESCC) and oral squamous cell carcinoma (OSCC), examining shared risk factors and underlying molecular mechanisms. By employing random forest (RF) classifier, enhanced with interpretable machine learning (IML) through SHapley Additive exPlanations (SHAP), we analyzed gene expression from two GEO datasets (GSE30784 and GSE44021). The GSE30784 dataset comprises 167 OSCC samples and 45 control group, whereas the GSE44021 dataset encompasses 113 ESCC samples and 113 control group. Our analysis led to identification of 20 key genes, such as XBP1, VGLL1, and RAD1, which are significantly associated with development of ESCC and OSCC. Further investigations were conducted using tools like NetworkAnalyst 3.0, Single Cell Portal, and miRNET 2.0, which highlighted complex interactions between these genes and specific miRNA targets including hsa-mir-124-3p and hsa-mir-1-3p. Our model achieved high precision in identifying genes linked to crucial processes like programmed cell death and cancer pathways, suggesting new avenues for diagnosis and treatment. This study confirms the bidirectional relationship between OSCC and ESCC, laying groundwork for targeted therapeutic approaches. This study helps to identify shared biological pathways and genetic factors of these conditions for designing personalized medicine strategies and to improve disease management.

Estimation of compressive strength of concrete with manufactured sand and natural sand using interpretable artificial intelligence

To satisfy the design strength of manufactured sand concrete (MSC) in practical engineering applications, a plethora of geotechnical tests are frequently conducted. An effective approach is imperative to reduce the consumption of labor and resources during these tests. The objective of this paper is to introduce an interpretable machine learning (ML) method to evaluate the compressive strength (CS) of MSC. Firstly, a dataset was established by compiling experimental results from 208 literatures. 3382 data points were selected from the dataset for algorithm training. Recursive Feature Elimination with Cross-Validation (RFECV) was employed to select input parameters. Four algorithms with 12 selected input variables and 1 output variable were constructed to predict CS of MSC using Random Forest (RF), Gradient Boosting Decision Trees (GBDT), eXtreme Gradient Boosting algorithm (XGBoost), and Categorical Boosting (CatBoost). The results show that XGBoost has the highest accuracy and generalization ability (R2=0.934, MAE=3.44, RMSE=5.16, MAPE=0.07). To enhance model transparency, SHapley Additive exPlanations (SHAP) was adopted to explain the underlying predictive mechanisms of ML models. Analyses show that, 1) the cement content, curing time, and water content were the main feature parameters influencing the CS of MSC, 2) the cement content, curing time, and water content have a linear increase, logarithmic increase, and exponential decrease with the CS of MSC, respectively. Partial Dependence Plots (PDP) and Individual Conditional Expectation (ICE) plots were used to further analyze the impacts of these significant influencing factors on the CS of MSC. Additionally, the Local Interpretable Model-Agnostic Explanations (LIME) method was employed to investigate thresholds for various material dosages in MSC containing 5–10 % stone powder. Two typical scenarios were selected for analysis, yielding recommended dosage ranges for concrete of two distinct strengths. Finally, a graphical user interface (GUI) for the CS of MSC has been designed, which might be of great use to material engineers. This provides reference and guidance for concrete engineering practice.

Interpretable machine learning models for predicting short-term prognosis in AChR-Ab+ generalized myasthenia gravis using clinical features and systemic inflammation index.

Myasthenia Gravis (MG) is an autoimmune disease that causes muscle weakness in 80% of patients, most of whom test positive for anti-acetylcholine receptor (AChR) antibodies (AChR-Abs). Predicting and improving treatment outcomes are necessary due to varying responses, ranging from complete relief to minimal improvement. Our study aims to develop and validate an interpretable machine learning (ML) model that integrates systemic inflammation indices with traditional clinical indicators. The goal is to predict the short-term prognosis (after 6 months of treatment) of AChR-Ab+ generalized myasthenia gravis (GMG) patients to guide personalized treatment strategies. We performed a retrospective analysis on 202 AChR-Ab+ GMG patients, dividing them into training and external validation cohorts. The primary outcome of this study was the Myasthenia Gravis Foundation of America (MGFA) post-intervention status assessed after 6 months of treatment initiation. Prognoses were classified as "unchanged or worse" for a poor outcome and "improved or better" for a good outcome. Accordingly, patients were categorized into "good outcome" or "poor outcome" groups. In the training cohort, we developed and internally validated various ML models using systemic inflammation indices, clinical indicators, or a combination of both. We then carried out external validation with the designated cohort. Additionally, we assessed the feature importance of our most effective model using the Shapley Additive Explanations (SHAP) method. In our study of 202 patients, 28.7% (58 individuals) experienced poor outcomes after 6 months of standard therapy. We identified 11 significant predictors, encompassing both systemic inflammation indexes and clinical metrics. The extreme gradient boosting (XGBoost) model demonstrated the best performance, achieving an area under the receiver operating characteristic (ROC) curve (AUC) of 0.944. This was higher than that achieved by logistic regression (Logit) (AUC: 0.882), random forest (RF) (AUC: 0.917), support vector machines (SVM) (AUC: 0.872). Further refinement through SHAP analysis highlighted five critical determinants-two clinical indicators and three inflammation indexes-as crucial for assessing short-term prognosis in AChR-Ab+ GMG patients. Our analysis confirms that the XGBoost model, integrating clinical indicators with systemic inflammation indexes, effectively predicts short-term prognosis in AChR-Ab+ GMG patients. This approach enhances clinical decision-making and improves patient outcomes.

Interpretable machine learning classifiers for the reliable prediction of fall induced hip fracture risk

Interpretable machine learning classifiers for the reliable prediction of fall induced hip fracture risk

Identification of texture MRI brain abnormalities on Fibromyalgia syndrome using interpretable machine learning models

To provide objective diagnostic markers for fibromyalgia symptoms (FMS) diagnosis, we have created interpretable extreme gradient boosting (XGBoost) models using radiomics to aid in the diagnosis of chronic pain (CP) and to develop nomogram models for diagnosing subgroups of FMS. A group of 54 patients with CP and 71 healthy controls was randomly separated into training and validation groups, using a 7:3 ratio. Radiomics features were extracted from grey-matter and white-matter in the filtered mwp0* image. The Mann-Whitney U test, Spearman’s rank correlation test, and least absolute shrinkage and selection operator (LASSO) were utilized to select features. An XGBoost model was created based on these features, and Shapley Additive exPlanations (SHAP) was used for personalization and visual interpretation. A nomogram was developed for the diagnosis of FMS subgroups, utilizing radiomics scores and clinical predictors. The efficacy of the nomogram was evaluated using the area under the receiver operating characteristic curve, while decision curve analysis was employed to evaluate its clinical efficacy. The XGBoost model displays stability in the training validation group, indicating lower overfitting of CP model. The nomogram model combined with the rad-score has a greater ability to distinguish between typical and sub-clinical than the clinical factor model alone. We developed and validated a CP diagnosis model by XGBoost and realized model visualization through SHAP. The rad-score obtained by machine learning was used to build a nomogram model that combines clinical scales to distinguish patients with typical and sub-clinical fibromyalgia.

Interpretable machine learning for predicting heavy metal removal efficiency in electrokinetic soil remediation

Electrokinetic remediation (EKR) presents a promising approach for polluted soil remediation, leveraging electric fields to mobilize contaminants and facilitate their removal. Accurate efficiency prediction is crucial for optimizing EKR processes, reducing costs, and minimizing environmental impact. However, the EKR efficiency is influenced by numerous complex factors, making it challenging to predict outcomes reliably. This study introduces a novel approach for developing interpretable machine learning (ML) models tailored for efficient and unbiased EKR efficiency prediction based on material properties and experimental conditions. A large experimental dataset comprising 185 tests was compiled, encompassing various EKR parameters, heavy metal concentration, soil characteristics. A structured ML workflow was devised, evaluating diverse ML algorithms and employing model-agnostic interpretation methodologies to uncover intrinsic correlations between removal efficiency and pertinent attributes. Among the investigated ML models, Extreme Gradient Boosting (XGB) exhibited the best performance. Bayesian optimization was applied to further enhance its capabilities. The optimized XGB model demonstrated superior predictive accuracy on the test set, with an RMSE of 0.255, an MAE of 0.158, and an R² of 0.82. Moreover, SHapley Additive exPlanations analysis was employed to interpret the contributions of the input variables. Among inputs, electrode distance emerged as the most significant factor influencing EKR efficiency, followed by area, electrolyte species, and remediation duration. This research not only advances the predictive capabilities for EKR efficiency but also provides crucial insights into the underlying factors influencing remediation success, paving the way for more efficient and scalable applications in environmental management.

Important structural parameter for curvature effect of TM-N4 embeded C70 fullerenes as electrocatalysts for CO2 reduction interpreted with machine learning and first-principles calculations

In this work, TM-N4 sites embedded on curved C70 Fullerenes (TMN4(n)@C64, n = 1–5 which denote the doping positions and TM = Sc-Zn) were designed as efficient catalysts for electrochemical CO2 reduction reaction (ECO2RR). With interpretable machine learning (ML) and first-principles calculations, the performance of TM-N4 sites as possible electrocatalysts for ECO2RR under the curvature effect were revealed. After the screen procedure of catalyst stability, CO2 activation and selectivity of 50 designed systems, VN4(3), CrN4(3), MnN4(2), FeN4(2, 3, 5)@C64 were identified as high performance catalysts for ECO2RR to CH4, with limiting potential of -0.41 V, -0.26 V, -0.28 V, -0.35 V, -0.25 V and -0.17 V respectively, outperform the TM-N4 catalysts on the flat surface. A key structural descriptor K which can reflect the curvature effect of TM-N4 sites on the curved surface was identified with ML models. The gradient boosting regression (GBR) and sure independence screening and sparsifying operator (SISSO) algorithm reveal that the structural parameter K have strong association with the CO2 adsorption and significant influence on the selectivity of catalysts. Interestingly, the structural parameter K can influence more the binding of O-bonded intermediates than C-bonded intermediates. With the GBR algorithm, an effective model for the prediction of the limiting potential for C1 products were obtained. This study clarified the curvature effect of TM-N4 supported on the curved surface, and provides valuable guidance for the designing of TM-N4 single atom catalysts for ECO2RR on curved surface.

Prediction and validation of pathologic complete response for locally advanced rectal cancer under neoadjuvant chemoradiotherapy based on a novel predictor using interpretable machine learning

BackgroundPrecise evaluation of pathological complete response (pCR) is essential for determining the prognosis of patients with locally advanced rectal cancer (LARC) undergoing neoadjuvant chemoradiotherapy (NCRT) and can offer clues for the selection of subsequent treatment strategies. Most current predictive models for pCR focus primarily on pre-treatment factors, neglecting the dynamic systemic changes that occur during neoadjuvant chemoradiotherapy, and are constrained by low accuracy and lack of integrity. PurposeThis study devised a novel predictor of pCR using dynamic alterations in systemic inflammation-nutritional marker indexes (SINI) during neoadjuvant therapy and developed a machine-learning model to predict pCR. MethodsTwo cohorts of patients with LARC from center one from 2012 to 2017 and from center two from 2020 to 2023 were integrated for analysis. This study compared dynamic changes in blood indexes before and after neoadjuvant therapy and surgical operation. A least absolute shrinkage and selection operator (LASSO) regression analysis was conducted to mitigate collinearity and identify key indexes, constructing the SINI. Univariate and multiple logistic regression analyses were used to identify the independent risk factors associated with pCR. Additionally, 10 machine learning algorithms were employed to develop predictive models to assess risk. The hyperparameters of the machine learning models were optimized using a random search and 10-fold cross-validation. The models were assessed by examining various metrics, including the area under the receiver operating characteristic curves (AUC), the area under the precision-recall curve (AUPRC), decision curve analysis, calibration curves, and the precision and accuracy of the internal and external validation cohorts. Additionally, Shapley's additive explanations (SHAP) were employed to interpret the machine learning models. ResultsThe study cohort comprised 677 patients from the center one and 224 patients from the center two. Six key indexes were identified, and a predictive index, SINI, was constructed. Univariate and multiple logistic regression analyses revealed that SINI, clinical T-stage, clinical N-stage, tumor size, and the distance from the anal verge were independent risk factors for pCR in patients with LARC following NCRT. The mean AUC value of the extreme gradient boosting (XGB) model in the 10-fold cross-validation of the training set was 0.877. The XGB model demonstrated superior performance in the internal and external validation sets. Specifically, in the internal test set, the XGB model achieved an AUC of 0.86, AUPRC of 0.707, accuracy of 0.82, and precision of 0.80. In the external validation set, the XGB model exhibited an AUC of 0.83, AUPRC of 0.702, accuracy of 0.81, and precision of 0.81. Additionally, the predictions generated by the XGB model were analyzed using SHAP. ConclusionThis study involved developing and validating an XGB model using SINI to predict pCR in patients with LARC. Besides, a SINI-based machine learning model shows promise in accurately predicting pCR following NCRT in patients with resectable LARC, offering valuable insights for personalized treatment approaches.

Rapid shear capacity prediction of TRM-strengthened unreinforced masonry walls through interpretable machine learning deployed in a web app

The presented study provides an efficient and reliable tool for rapid shear capacity estimation of TRM-strengthened unreinforced masonry walls. For this purpose, a data-driven methodology based on a machine learning system is proposed using a dataset constituted of experimental results selected from the bibliography. The outlier points of the dataset were detected using the Cook’s distance methodology and removed from the raw dataset, which consisted of 113 examples and 11 input variables. In the processed dataset, 17 machine learning methods were trained, optimized through hyperparameter tuning, and compared on the test set. The most effective models were the optimized instances of XGBoost and CatBoost methods, which combined into a voting model to leverage the predictive capacity of more than a single regressor. The final blended model shows remarkable predicting capacity with the determination factor (R2) equal to 0.95 and the mean absolute percentage error equal to 8.03%. Also, the model’s predictions are compared with those of existing analytical relationships, and it is found to perform the best of all. In sequence, machine learning interpretation methods are applied to find how the predictors influence the target output. Am, ft, and n⋅tf were identified as the most significant predictors with a mainly positive influence on the shear capacity. Finally, the built machine learning system is employed in a user-friendly web app for easy access and usage by professionals and researchers.