SHapley Additive exPlanations Values Research Articles

Foodinsecurity.london: Developing a food-insecurity prevalence map for London - a machine learning from food-sharing footprints

Introduction & BackgroundThe ability of policymakers to positively transform food environments requires robust empirical evidence that can inform decisions. At present, there is limited data on food-insecurity in the UK that can be used to inform interventions by local authorities, due to the prohibitive costs and logistical challenges of administering longitudinal surveys. This study builds on existing research and a key pilot study developed in partnership between Olio - a food-sharing app with 7 million registered users as of 2023, the University of Nottingham and Havering Council in 2020, which resulted in the world’s first map prototype of food-insecurity. Objectives & ApproachOur approach leverages Machine Learning methods applied to unprecedented food-acquisition behavioural data and open area-level deprivation statistics to model and predict individuals' experience of food-insecurity across London. We used Olio’s extensive network of users to distribute 2,849 surveys, asking respondents across London about their experiences of food-insecurity. The survey was distributed online, adapting the US Department of Agriculture Food Security module. Respondents were asked about their experiences, including (1) eating smaller meals or skipping meals, (2) being hungry but being unable to eat, and (3) not eating for a whole day, because they could not afford food or because they could not get access to food. Using the household, rather than the individual-level version of the food insecurity module helped shed light on the experience of vulnerable groups - such as children. Relevance to Digital FootprintsThe survey responses provided a ground truth about users' experiences of destitution. Deprivation metrics and digital footprint data in the form of food-acquisition behavioural data were then used in a Random Forests Machine Learning model to predict whether households were experiencing food-insecurity, achieving high accuracy. Food-sharing data from almost 50,000 London-based users active on Olio’s platform were then used to identify relevant food-seeking behaviours and aggregate recognised instances of food-insecurity at neighbourhood (MSOA) level. Conclusions & ImplicationsTo identify and rank relevant socio-demographics and food-seeking behaviours most informative for describing food-insecurity an extensive variable selection analysis was performed. The resulting SHAP (SHapley Additive exPlanations) values showed that a combination of food solicitation and the general deprivation of an area were important predictors of food-insecurity.

A non-linear and interaction effect analysis of various risk factors influencing mobile phone use while driving among long-haul truck drivers travelling across India

Mobile phone use while driving (MPWD) is one of the most critical behaviours contributing to distraction related fatalities and injuries. However, MPWD is not well understood among long-haul truck drivers (LHTDs) especially in developing countries like India. The present study aims to develop a prediction model for MPWD and attempts to identify the underlying factors and patterns affecting MPWD among LHTDs travelling across India. A total of 756 valid samples were collected utilising a questionnaire in Salem city, Tamil Nadu, India. Machine learning algorithms including Decision tree, Random Forest, Adaptive Boosting, and Extreme gradient boosting were employed to model MPWD. A split ratio of 70:30 was adopted for training and testing purposes. A 10-fold cross validation procedure was carried out during model development. The analysis results showed that XGBoost demonstrated superior performance than other models (accuracy = 0.82, F1 score =0.8, and AUROC = 0.82). In addition, SHapley Additive explanations (SHAP) were implemented to reveal the importance of predictors contributing to MPWD. The findings of the study showed that type of commodity, pressured delivery, calls received during driving, smoking habits, educational level, and continuous driving duration as major factors contributing to MPWD among LHTDs. SHAP values were also utilised to investigate hidden patterns and interaction effects for major factors affecting MPWD. Parcel and food items coupled with frequent pressured delivery and middle aged LHTDs driving continuously for 5–6 h are some of the interactive risk factors enhancing MPWD. These findings are useful for road safety authorities and truck companies to draft suitable policies and help to introduce effective interventions for reducing MPWD among LHTDs.

Prediction and Rational Design of Stacking Fault Energy of Austenitic Alloys Based on Interpretable Machine Learning and Chemical Composition

Accurately predicting the stacking fault energy (SFE), as one of the crucial factors influencing the material deformation mechanism, is a focal point in research. This study utilizes measured SFE values from the literature on austenitic alloys to establish a predictive model for the relationship between chemical composition and SFE using machine learning techniques. Among five compared machine learning algorithms, the extremely randomized trees algorithm demonstrates the highest prediction accuracy. Incorporating atomic features further enhances the model's performance. Subsequently, feature selection is conducted using correlation analysis, recursive elimination, and exhaustive search to obtain the optimal subset of features. The interpretability of the model is analyzed using Shapley additive explanations values, providing rankings of feature importance and critical thresholds for feature influence. Finally, rational design of SFE is demonstrated by adjusting alloy composition, exemplified by Fe–17.5Cr–15.6Ni–2.5Mo–0.02Si–0Co steel.

Fracture risk prediction in postmenopausal women with traditional and machine learning models in a nationwide, prospective cohort study in Switzerland with validation in the UK Biobank.

Fracture prediction is essential in managing patients with osteoporosis and is an integral component of many fracture prevention guidelines. We aimed to identify the most relevant clinical fracture risk factors in contemporary populations by training and validating short- and long-term fracture risk prediction models in 2 cohorts. We used traditional and machine learning survival models to predict risks of vertebral, hip, and any fractures on the basis of clinical risk factors, T-scores, and treatment history among participants in a nationwide Swiss Osteoporosis Registry (N = 5944 postmenopausal women, median follow-up of 4.1yr between January 2015 and October 2022; a total of 1190 fractures during follow-up). The independent validation cohort comprised 5474 postmenopausal women from the UK Biobank with 290 incident fractures during follow-up. Uno's C-index and the time-dependent area under the receiver operating characteristics curve were calculated to evaluate the performance of different machine learning models (Random survival forest and eXtreme Gradient Boosting). In the independent validation set, the C-index was 0.74 [0.58, 0.86] for vertebral fractures, 0.83 [0.7, 0.94] for hip fractures, and 0.63 [0.58, 0.69] for any fractures at year 2, and these values further increased for longer estimations of up to 7yr. In comparison, the 10-yr fracture probability calculated with FRAX Switzerland was 0.60 [0.55, 0.64] for major osteoporotic fractures and 0.62 [0.49, 0.74] for hip fractures. The most important variables identified with Shapley additive explanations values were age, T-scores, and prior fractures, while number of falls was an important predictor of hip fractures. Performances of both traditional and machine learning models showed similar C-indices. We conclude that fracture risk can be improved by including the lumbar spine T-score, trabecular bone score, numbers of falls and recent fractures, and treatment information has a significant impact on fracture prediction.

Optimizing Lithium-Ion Battery Performance: Integrating Machine Learning and Explainable AI for Enhanced Energy Management

Managing the capacity of lithium-ion batteries (LiBs) accurately, particularly in large-scale applications, enhances the cost-effectiveness of energy storage systems. Less frequent replacement or maintenance of LiBs results in cost savings in the long term. Therefore, in this study, AdaBoost, gradient boosting, XGBoost, LightGBM, CatBoost, and ensemble learning models were employed to predict the discharge capacity of LiBs. The prediction performances of each model were compared based on mean absolute error (MAE), mean squared error (MSE), and R-squared values. The research findings reveal that the LightGBM model exhibited the lowest MAE (0.103) and MSE (0.019) values and the highest R-squared (0.887) value, thus demonstrating the strongest correlation in predictions. Gradient boosting and XGBoost models showed similar performance levels but ranked just below LightGBM. The competitive performance of the ensemble model indicates that combining multiple models could lead to an overall performance improvement. Furthermore, the study incorporates an analysis of key features affecting model predictions using SHAP (Shapley additive explanations) values within the framework of explainable artificial intelligence (XAI). This analysis evaluates the impact of features such as temperature, cycle index, voltage, and current on predictions, revealing a significant effect of temperature on discharge capacity. The results of this study emphasize the potential of machine learning models in LiB management within the XAI framework and demonstrate how these technologies could play a strategic role in optimizing energy storage systems.

A retrospective machine learning–based analysis of nationwide cancer comprehensive genomic profiling data to identify features associated with recommendation of mutation-based therapy.

e13510 Background: Comprehensive genomic profiling (CGP) has played key roles in cancer precision medicine through optimization of therapeutic interventions based on genomic alterations in cancer cells. However, the probability of discovering mutation-based treatments through CGP remains low. To enhance the effectiveness and efficiency of cancer precision medicine, it is crucial to identify patients who are likely to benefit from CGP tests. This study aims to identify characteristics of patients in which mutation-based treatments are discovered by CGP tests. Methods: We retrospectively analyzed data from 60,655 patients who underwent CGP tests and were registered in the Center for Cancer Genomics and Advanced Therapeutics (C-CAT), the national data center for cancer CGP in Japan. The C-CAT database covers 99.7% of cancer patients who have undergone CGP tests in Japan. Major cancer types included 10,182 cases of bowel cancer, 8,691 pancreas cancer, 5,062 biliary tract cancer, and 3,777 breast cancer. We developed an eXtreme Gradient Boosting (XGBoost) model using machine learning, and used clinical information as input to predict whether one or more Japanese Pharmaceuticals and Medical Devices Agency (PMDA)-approved drugs are discovered through CGP tests or not. Shapley Additive Explanations (SHAP) was employed to extract significant features that contribute to the model prediction. Results: The prediction model achieved an area under the receiver operating characteristic curve of 0.826 for the overall cancer population. Positive SHAP values were observed for patients with breast (mean SHAP in breast cancer patients: 1.38), lung (1.08), bowel (0.75), and pancreas cancers (0.35), while negative SHAP values were associated with head and neck (-1.75), cervical cancers (-1.54), and brain (-1.33). Positive SHAP values were also associated with presence of liver or lymph node metastasis (0.23, 0.08), shorter intervals between diagnosis and CGP testing or specimen collection and CGP testing, and advanced age. Similar trends were observed in cancer-type-specific prediction models, which also identified their own unique features. In the adolescent and young adult (AYA) age group, primary brain tumors were strongly associated with negative SHAP values (-1.37). Conclusions: Our machine learning-based analysis of nationwide CGP data identified features that predict cases in which mutation-based treatments are discovered by CGP tests, both in the overall cancer population and within specific cancer types and the AYA age group. Expedited CGP testing is recommended for patients who match the identified profile to facilitate early targeted therapy interventions.

Unleashing deep features and radiomics to enhance best overall response prediction to immunotherapy in advanced non-small cell lung cancer.

e20611 Background: Predicting immunotherapy response in advanced non-small cell lung cancer (NSCLC) through non-invasive means could be a groundbreaking advancement. Deep features extracted from medical images by deep Convolutional Neural Networks (CNNs) along with radiomics have demonstrated promising results in different oncologic diseases. However, the method for extracting these deep features influences the predictive capabilities of the model. Methods: A retrospective analysis of a multicenter study was conducted in advanced NSCLC patients with high PD-L1 expression with the aim of developing an imaging features-based model for the prediction of best overall response (BOR) to a checkpoint inhibitor immunotherapy. All visible lung lesions, beyond RECIST 1.1, were manually segmented in baseline CT scans. Radiomics features were extracted at lesion-level, and various preprocessing methods and CNN architectures were tested for deep features extraction. Radiomics and deep features were harmonized using ComBat methodology. A distinctive feature signature was created for each patient by calculating the mean across all lesions. Using this data, machine learning models were trained to predict treatment response, specifically distinguishing between complete response (CR) and partial response (PR) from stable disease (SD) and progressive disease (PD). The study compared the performance of different deep features extraction methods and assessed shapley additive explanations (SHAP) values of radiomics for interpretability. Results: In the cohort of 116 patients (mean age: 67, 93% ever-smokers, 73% male), 1024 to 5475 quantitative imaging features per lesion were extracted. Among these patients, 67 patients were CR+PR, and 49 were SD+PD. The top-performing predictive model, trained with radiomics and deep features from a deep convolutional autoencoder, achieved an average AUC of 85%. SHAP values showed that impactful radiomics features were correlated with intensity changes in adjacent pixels describing the tumor heterogeneity, along with shape features potentially associated with malignancy. Ultimately, combining both features demonstrated statistical significance, leading to a performance improvement of ~2-5% compared to models exclusively using deep features. Conclusions: Diverse techniques for extracting deep features were evaluated, revealing autoencoder training as the most effective strategy for predicting immunotherapy response. Consequently, the findings suggest that employing deep learning to extract imaging features from baseline CT scans, combined with radiomics, represents a promising non-invasive approach for the predicting immunotherapy outcomes in advanced NSCLC patients.

P‐6: Redefining Pixel Circuit Analysis: Causal Discovery and Probabilistic Modeling

When analyzing data using existing machine learning models without explicit causal information, several limitations often arise, particularly in the misinterpretation of correlations as causal relationships. These limitations are more pronounced in complex scenarios or in situations where outcomes are influenced by sequential effects. This paper presents an advanced methodology for analyzing pixel circuit design and its driving conditions, a domain characterized by complex interactions and multiple variables. Traditional machine learning methods, when applied in isolation, have shown limitations in unraveling the intricate causal relationships inherent in such systems. To address this challenge, we integrated Explainable AI (XAI) techniques, particularly SHAP (SHapley Additive exPlanations) values, into our analysis. In collaboration with domain experts, we constructed a Directed Acyclic Graph (DAG) that effectively reduced the complexity of interconnections and ensured consistency with empirical data. This approach facilitated a more accurate identification of the impact of each parameter and its causal influence. By decomposing the joint distribution of the variables into conditional distributions, taking into account their parental relationships, we gained a deeper understanding of the changes in causal mechanisms. Our methodology significantly enhanced the accuracy of causal analysis under realistic pixel driving conditions. The findings not only offer novel insights into pixel circuitry but also demonstrate the efficacy of combining machine learning with XAI in complex systems analysis, gaining wide acceptability among relevant experts.

Determining the Climatic Drivers for Wine Production in the Côa Region (Portugal) Using a Machine Learning Approach

The Côa region in inner-northern Portugal heavily relies on viticulture, which is a cornerstone of its economy and cultural identity. Understanding the intricate relationship between climatic variables and wine production (WP) is crucial for adapting management practices to changing climatic conditions. This study employs machine learning (ML), specifically random forest (RF) regression, to predict grapevine yields in the Côa region using high-resolution climate data for 2004–2020. SHAP (SHapley Additive exPlanations) values are used to potentially explain the non-linear relationships between climatic factors and WP. The results reveal a complex interplay between predictors and WP, with precipitation emerging as a key determinant. Higher precipitation levels in April positively impact WP by replenishing soil moisture ahead of flowering, while elevated precipitation and humidity levels in August have a negative effect, possibly due to late-season heavy rainfall damaging grapes or creating more favorable conditions for fungal pathogens. Moreover, warmer temperatures during the growing season and adequate solar radiation in winter months favor higher WP. However, excessive radiation during advanced growth stages can lead to negative effects, such as sunburn. This study underscores the importance of tailoring viticultural strategies to local climatic conditions and employing advanced analytical techniques such as SHAP values to interpret ML model predictions effectively. Furthermore, the research highlights the potential of ML models in climate change risk reduction associated with viticulture, specifically WP. By leveraging insights from ML and interpretability techniques, policymakers and stakeholders can develop adaptive strategies to safeguard viticultural livelihoods and stable WP in a changing climate, particularly in regions with a rich agrarian heritage, such as the Côa region.

Machine learning prediction on wetland succession and the impact of artificial structures from a decade of field data

The artificial structures can influence wetland topology and sediment properties, thereby shaping plant distribution and composition. Macrobenthos composition was correlated with plant cover. Previous studies on the impact of artificial structures on plant distribution are scarce in incorporating time-series data or extended field surveys. In this study, a machine-learning-based species distribution model with decade-long observation was analyzed to investigate the correlation between the shift in the distribution of B. planiculmis, artificial structure-induced elevation changes and the expansion of other plants, as well as their connection to soil properties and crab composition dynamics under plants in Gaomei Wetland. Long short-term memory model (LSTM) with Shapley additive explanations (SHAP) was employed for predicting the distribution of B. planiculmis and explaining feature importance. The results indicated that wetland topology was influenced by both artificial structures and plants. Areas initially colonized by B. planiculmis were replaced by other species. Soil properties showed significant differences among plant patches; however, principal component analysis (PCA) of sediment properties and niche similarity analysis showed that the niche of plants was overlapped. Crab composition was different under different plants. The presence probability of B. planiculmis near woody paths decreased according to LSTM and field survey data. SHAP analysis suggested that the distribution of other plants, historical distribution of B. planiculmis and sediment properties significantly contributed to the presence probability of B. planiculmis. A sharp decrease in SHAP values with increasing NDVI at suitable elevations, overlap in PCA of sediment properties and niche similarity indicated potential competition among plants. This decade-long time-series field survey revealed the joint effects of artificial structure and vegetation on the topology and soil properties dynamics. These changes influenced the plant distribution through potential plant competition. LSTM with SHAP provided valuable insights in the underlying the mechanisms of artificial structure effects on the plant zonation process.

Enhancing hERG Risk Assessment with Interpretable Classificatory and Regression Models.

The human Ether-à-go-go-Related Gene (hERG) is a transmembrane protein that regulates cardiac action potential, and its inhibition can induce a potentially deadly cardiac syndrome. In vitro tests help identify hERG blockers at early stages; however, the high cost motivates searching for alternative, cost-effective methods. The primary goal of this study was to enhance the Pred-hERG tool for predicting hERG blockage. To achieve this, we developed new QSAR models that incorporated additional data, updated existing classificatory and multiclassificatory models, and introduced new regression models. Notably, we integrated SHAP (SHapley Additive exPlanations) values to offer a visual interpretation of these models. Utilizing the latest data from ChEMBL v30, encompassing over 14,364 compounds with hERG data, our binary and multiclassification models outperformed both the previous iteration of Pred-hERG and all publicly available models. Notably, the new version of our tool introduces a regression model for predicting hERG activity (pIC50). The optimal model demonstrated an R2 of 0.61 and an RMSE of 0.48, surpassing the only available regression model in the literature. Pred-hERG 5.0 now offers users a swift, reliable, and user-friendly platform for the early assessment of chemically induced cardiotoxicity through hERG blockage. The tool provides versatile outcomes, including (i) classificatory predictions of hERG blockage with prediction reliability, (ii) multiclassificatory predictions of hERG blockage with reliability, (iii) regression predictions with estimated pIC50 values, and (iv) probability maps illustrating the contribution of chemical fragments for each prediction. Furthermore, we implemented explainable AI analysis (XAI) to visualize SHAP values, providing insights into the contribution of each feature to binary classification predictions. A consensus prediction calculated based on the predictions of the three developed models is also present to assist the user's decision-making process. Pred-hERG 5.0 has been designed to be user-friendly, making it accessible to users without computational or programming expertise. The tool is freely available at http://predherg.labmol.com.br.

Explainable AI models for predicting liquefaction-induced lateral spreading

Introduction: Earthquake-induced liquefaction can cause substantial lateral spreading, posing threats to infrastructure. Machine learning (ML) can improve lateral spreading prediction models by capturing complex soil characteristics and site conditions. However, the “black box” nature of ML models can hinder their adoption in critical decision-making.Method: This study addresses this limitation by using SHapley Additive exPlanations (SHAP) to interpret an eXtreme Gradient Boosting (XGB) model for lateral spreading prediction, trained on data from the 2011 Christchurch Earthquake.Result: SHAP analysis reveals the factors driving the model's predictions, enhancing transparency and allowing for comparison with established engineering knowledge. Notably, the SHAP values expose an unexpected behavior in the PGA feature. Moreover, the results demonstrate that the XGB model successfully identifies the importance of soil characteristics derived from Cone Penetration Test (CPT) data in predicting lateral spreading, validating its alignment with domain understanding.Discussion: This work highlights the value of explainable machine learning for reliable and informed decision-making in geotechnical engineering and hazard assessment.

Towards proactive palliative care in oncology: developing an explainable EHR-based machine learning model for mortality risk prediction

BackgroundEx-ante identification of the last year in life facilitates a proactive palliative approach. Machine learning models trained on electronic health records (EHR) demonstrate promising performance in cancer prognostication. However, gaps in literature include incomplete reporting of model performance, inadequate alignment of model formulation with implementation use-case, and insufficient explainability hindering trust and adoption in clinical settings. Hence, we aim to develop an explainable machine learning EHR-based model that prompts palliative care processes by predicting for 365-day mortality risk among patients with advanced cancer within an outpatient setting.MethodsOur cohort consisted of 5,926 adults diagnosed with Stage 3 or 4 solid organ cancer between July 1, 2017, and June 30, 2020 and receiving ambulatory cancer care within a tertiary center. The classification problem was modelled using Extreme Gradient Boosting (XGBoost) and aligned to our envisioned use-case: “Given a prediction point that corresponds to an outpatient cancer encounter, predict for mortality within 365-days from prediction point, using EHR data up to 365-days prior.” The model was trained with 75% of the dataset (n = 39,416 outpatient encounters) and validated on a 25% hold-out dataset (n = 13,122 outpatient encounters). To explain model outputs, we used Shapley Additive Explanations (SHAP) values. Clinical characteristics, laboratory tests and treatment data were used to train the model. Performance was evaluated using area under the receiver operating characteristic curve (AUROC) and area under the precision-recall curve (AUPRC), while model calibration was assessed using the Brier score.ResultsIn total, 17,149 of the 52,538 prediction points (32.6%) had a mortality event within the 365-day prediction window. The model demonstrated an AUROC of 0.861 (95% CI 0.856–0.867) and AUPRC of 0.771. The Brier score was 0.147, indicating slight overestimations of mortality risk. Explanatory diagrams utilizing SHAP values allowed visualization of feature impacts on predictions at both the global and individual levels.ConclusionOur machine learning model demonstrated good discrimination and precision-recall in predicting 365-day mortality risk among individuals with advanced cancer. It has the potential to provide personalized mortality predictions and facilitate earlier integration of palliative care.

Machine Learning-Based Prediction of Suicidal Thinking in Adolescents by Derivation and Validation in 3 Independent Worldwide Cohorts: Algorithm Development and Validation Study.

Suicide is the second-leading cause of death among adolescents and is associated with clusters of suicides. Despite numerous studies on this preventable cause of death, the focus has primarily been on single nations and traditional statistical methods. This study aims to develop a predictive model for adolescent suicidal thinking using multinational data sets and machine learning (ML). We used data from the Korea Youth Risk Behavior Web-based Survey with 566,875 adolescents aged between 13 and 18 years and conducted external validation using the Youth Risk Behavior Survey with 103,874 adolescents and Norway's University National General Survey with 19,574 adolescents. Several tree-based ML models were developed, and feature importance and Shapley additive explanations values were analyzed to identify risk factors for adolescent suicidal thinking. When trained on the Korea Youth Risk Behavior Web-based Survey data from South Korea with a 95% CI, the XGBoost model reported an area under the receiver operating characteristic (AUROC) curve of 90.06% (95% CI 89.97-90.16), displaying superior performance compared to other models. For external validation using the Youth Risk Behavior Survey data from the United States and the University National General Survey from Norway, the XGBoost model achieved AUROCs of 83.09% and 81.27%, respectively. Across all data sets, XGBoost consistently outperformed the other models with the highest AUROC score, and was selected as the optimal model. In terms of predictors of suicidal thinking, feelings of sadness and despair were the most influential, accounting for 57.4% of the impact, followed by stress status at 19.8%. This was followed by age (5.7%), household income (4%), academic achievement (3.4%), sex (2.1%), and others, which contributed less than 2% each. This study used ML by integrating diverse data sets from 3 countries to address adolescent suicide. The findings highlight the important role of emotional health indicators in predicting suicidal thinking among adolescents. Specifically, sadness and despair were identified as the most significant predictors, followed by stressful conditions and age. These findings emphasize the critical need for early diagnosis and prevention of mental health issues during adolescence.

Pain prediction model based on machine learning and SHAP values for elders with dementia in Taiwan

IntroductionPain conditions are common in elderly individuals, including those with dementia. However, symptoms associated with dementia may lead to poor recognition, assessment and management of pain. In this study, we incorporated the variables based on questionnaires into a machine learning algorithm to build a prediction model for the pain index of elderly individuals with dementia. Materials and methodsIn this study, 113 cases were collected through questionnaires and used to build prediction models for the patient's pain index. Three machine learning algorithms were incorporated for comparison in this study. To interpret the prediction model, SHapley additive explanations values were used to depict the ranking importance of variables and the relationship between features and pain index. ResultsIn the comparison of models, random forests with feature selection outperformed in terms of root mean square error and mean absolute error. A total of 11 features were selected based on embedded method. The results showed that the Karnofsky scale played a key role in predicting pain index for elderly individuals with dementia and was positively associated with pain index. Arthritis is the most important disease to predicting the pain index. ConclusionsOur findings provided the key insights to predict the pain index of elderly patients with dementia. In the future, it can be used to develop an application system or webpage, which can reduce the use of labour and improve the efficiency.

Creating machine learning models that interpretably link systemic inflammatory index, sex steroid hormones, and dietary antioxidants to identify gout using the SHAP (SHapley Additive exPlanations) method.

The relationship between systemic inflammatory index (SII), sex steroid hormones, dietary antioxidants (DA), and gout has not been determined. We aim to develop a reliable and interpretable machine learning (ML) model that links SII, sex steroid hormones, and DA to gout identification. The dataset we used to study the relationship between SII, sex steroid hormones, DA, and gout was from the National Health and Nutrition Examination Survey (NHANES). Six ML models were developed to identify gout by SII, sex steroid hormones, and DA. The seven performance discriminative features of each model were summarized, and the eXtreme Gradient Boosting (XGBoost) model with the best overall performance was selected to identify gout. We used the SHapley Additive exPlanation (SHAP) method to explain the XGBoost model and its decision-making process. An initial survey of 20,146 participants resulted in 8,550 being included in the study. Selecting the best performing XGBoost model associated with SII, sex steroid hormones, and DA to identify gout (male: AUC: 0.795, 95% CI: 0.746- 0.843, accuracy: 98.7%; female: AUC: 0.822, 95% CI: 0.754- 0.883, accuracy: 99.2%). In the male group, The SHAP values showed that the lower feature values of lutein + zeaxanthin (LZ), vitamin C (VitC), lycopene, zinc, total testosterone (TT), vitamin E (VitE), and vitamin A (VitA), the greater the positive effect on the model output. In the female group, SHAP values showed that lower feature values of E2, zinc, lycopene, LZ, TT, and selenium had a greater positive effect on model output. The interpretable XGBoost model demonstrated accuracy, efficiency, and robustness in identifying associations between SII, sex steroid hormones, DA, and gout in participants. Decreased TT in males and decreased E2 in females may be associated with gout, and increased DA intake and decreased SII may reduce the potential risk of gout.

Cross-Platform Bug Localization Strategies: Utilizing Machine Learning for Diverse Software Environment Adaptability

This paper introduces a novel hybrid machine learning model that combines Long Short-Term Memory (LSTM) networks and SHapley Additive exPlanations (SHAP) to enhance bug localization across multiple software platforms. The aim is to adapt to the variability inherent in different operating systems and provide transparent, interpretable results for software developers. Our methodology includes comprehensive preprocessing of bug report data using advanced natural language processing techniques, followed by feature extraction through word embeddings to accommodate the sequential nature of text data. The LSTM model is trained and evaluated on a dataset of simulated bug reports, with the results interpreted using SHAP values to ensure clarity in decision-making. The results demonstrate the model’s robustness, adaptability, and consistent performance across platforms, as evidenced by accuracy, precision, recall, and F1 scores. The dataset's distribution of bug categories and statuses further provides valuable insights into common software development issues.

An exploratory framework for mapping, mechanism, and management of urban soundscape quality: From quietness to naturalness

BackgroundDespite growing attention from researchers and governments, challenges persist in comprehensively assessing urban sound quality by integrating both quietness and naturalness aspects. GoalsThis study aimed to develop an innovative soundscape quality index that concurrently evaluates quietness and naturalness in urban soundscapes. Our objectives included conducting urban soundscape quality mapping, analyzing influential mechanisms, and identifying priority zones for sound environment management. ApproachesWe collected sound pressure level (SPL) and raw audio data, from which we computed a normalized difference soundscape index (NDSI). With a dataset comprising 28 explanatory variables encompassing land use, built environment, vegetation characteristics, and temporal factors, we employed the random forest (RF) model to predict 10 indicators, including eight SPL-related indices, NDSI, and the QNS (quietness and naturalness soundscape) index. Crucially, we utilized SHAP (SHapley Additive exPlanations) values to interpret the RF model. FindingsSpatial variations in quietness and naturalness were evident, closely associated with road networks and vegetation, respectively, with discernible temporal variations. The top three variables influencing QNS were distance to major roads, normalized difference vegetation index (NDVI), and proportion of tree coverage. Moreover, interaction effects highlighted dual negative or synergistic promoting effects on QNS from factors such as road width, human disturbance, vegetation configurations, and land cover. Notably, these mechanisms were successfully applied to six typical tourist attractions in Xiamen city, where five types of management zones were mapped based on priority considerations of population density and soundscape quality. Interestingly, natural soundscape reserves were highly correlated with city parks, high-risk zones predominantly overlapped with road networks, and potential zones comprised inner communities between streets. SignificanceThe framework demonstrated effectiveness in mapping, exploring mechanisms, and guiding management strategies for the urban sound environment.

AI for Automating Data Center Operations: Model Explainability in the Data Centre Context Using Shapley Additive Explanations (SHAP)

The application of Artificial Intelligence (AI) and Machine Learning (ML) models is increasingly leveraged to automate and optimize Data Centre (DC) operations. However, the interpretability and transparency of these complex models pose critical challenges. Hence, this paper explores the Shapley Additive exPlanations (SHAP) values model explainability method for addressing and enhancing the critical interpretability and transparency challenges of predictive maintenance models. This method computes and assigns Shapley values for each feature, then quantifies and assesses their impact on the model’s output. By quantifying the contribution of each feature, SHAP values can assist DC operators in understanding the underlying reasoning behind the model’s output in order to make proactive decisions. As DC operations are dynamically changing, we additionally investigate how SHAP can capture the temporal behaviors of feature importance in the dynamic DC environment over time. We validate our approach with selected predictive models using an actual dataset from a High-Performance Computing (HPC) DC sourced from the Enea CRESCO6 cluster in Italy. The experimental analyses are formalized using summary, waterfall, force, and dependency explanations. We delve into temporal feature importance analysis to capture the features’ impact on model output over time. The results demonstrate that model explainability can improve model transparency and facilitate collaboration between DC operators and AI systems, which can enhance the operational efficiency and reliability of DCs by providing a quantitative assessment of each feature’s impact on the model’s output.

Prediction models for postoperative recurrence of non-lactating mastitis based on machine learning

ObjectivesThis study aims to build a machine learning (ML) model to predict the recurrence probability for postoperative non-lactating mastitis (NLM) by Random Forest (RF) and XGBoost algorithms. It can provide the ability to identify the risk of NLM recurrence and guidance in clinical treatment plan.MethodsThis study was conducted on inpatients who were admitted to the Mammary Department of Shuguang Hospital affiliated to Shanghai University of Traditional Chinese Medicine between July 2019 to December 2021. Inpatient data follow-up has been completed until December 2022. Ten features were selected in this study to build the ML model: age, body mass index (BMI), number of abortions, presence of inverted nipples, extent of breast mass, white blood cell count (WBC), neutrophil to lymphocyte ratio (NLR), albumin-globulin ratio (AGR) and triglyceride (TG) and presence of intraoperative discharge. We used two ML approaches (RF and XGBoost) to build models and predict the NLM recurrence risk of female patients. Totally 258 patients were randomly divided into a training set and a test set according to a 75%-25% proportion. The model performance was evaluated based on Accuracy, Precision, Recall, F1-score and AUC. The Shapley Additive Explanations (SHAP) method was used to interpret the model.ResultsThere were 48 (18.6%) NLM patients who experienced recurrence during the follow-up period. Ten features were selected in this study to build the ML model. For the RF model, BMI is the most important influence factor and for the XGBoost model is intraoperative discharge. The results of tenfold cross-validation suggest that both the RF model and the XGBoost model have good predictive performance, but the XGBoost model has a better performance than the RF model in our study. The trends of SHAP values of all features in our models are consistent with the trends of these features’ clinical presentation. The inclusion of these ten features in the model is necessary to build practical prediction models for recurrence.ConclusionsThe results of tenfold cross-validation and SHAP values suggest that the models have predictive ability. The trend of SHAP value provides auxiliary validation in our models and makes it have more clinical significance.