Interpretable Machine Learning Models Research Articles

Optimisation and interpretation of machine and deep learning models for improved water quality management in Lake Loktak

Loktak Lake, one of the largest freshwater lakes in Manipur, India, is critical for the eco-hydrology and economy of the region, but faces deteriorating water quality due to urbanisation, anthropogenic activities, and domestic sewage. Addressing the urgent need for effective pollution management, this study aims to assess the lake's water quality status using the water quality index (WQI) and develop advanced machine learning (ML) tools for WQI assessment and ML model interpretation to improve pollution management decision making. The WQI was assessed using entropy-based weighting arithmetic and three ML models — Gradient Boosting Machine (GBM), Random Forest (RF) and Deep Neural Network (DNN) — were optimised using a grid search algorithm in the H2O Application Programming Interface (API). These models were validated by various metrics and interpreted globally and locally via Partial Dependency Plot (PDP), Accumulated Local Effect (ALE) and SHapley Additive exPlanations (SHAP). The results show a WQI range of 72.38–100, with 52.7% of samples categorised as very poor. The RF model outperformed GBM and DNN and showed the highest accuracy and generalisation ability, which is reflected in the superior R2 values (0.97 in training, 0.9 in test) and the lower root mean square error (RMSE). RF's minimal margin of error and reliable feature interpretation contrasted with DNN's larger margin of error and inconsistency, which affected its usefulness for decision making. Turbidity was found to be a critical predictive feature in all models, significantly influencing WQI, with other variables such as pH and temperature also playing an important role. SHAP dependency plots illustrated the direct relationship between key water quality parameters such as turbidity and WQI predictions. The novelty of this study lies in its comprehensive approach to the evaluation and interpretation of ML models for WQI estimation, which provides a nuanced understanding of water quality dynamics in Loktak Lake. By identifying the most effective ML models and key predictive functions, this study provides invaluable insights for water quality management and paves the way for targeted strategies to monitor and improve water quality in this vital freshwater ecosystem.

Development of interpretable machine learning models associated with environmental chemicals to predict all-cause and specific-cause mortality：A longitudinal study based on NHANES

Limited information is available on potential predictive value of environmental chemicals for mortality. Our study aimed to investigate the associations between 43 of 8 classes representative environmental chemicals in serum/urine and mortality, and further develop the interpretable machine learning models associated with environmental chemicals to predict mortality. A total of 1602 participants were included from the National Health and Nutrition Examination Survey (NHANES). During 154,646 person-months of follow-up, 127 deaths occurred. We found that machine learning showed promise in predicting mortality. CoxPH was selected as the optimal model for predicting all-cause mortality with time-dependent AUROC of 0.953 (95%CI: 0.951–0.955). Coxnet was the best model for predicting cardiovascular disease (CVD) and cancer mortality with time-dependent AUROCs of 0.935 (95%CI: 0.933–0.936) and 0.850 (95%CI: 0.844–0.857). Based on clinical variables, adding environmental chemicals could enhance the predictive ability of cancer mortality (P < 0.05). Some environmental chemicals contributed more to the models than traditional clinical variables. Combined the results of association and prediction models by interpretable machine learning analyses, we found urinary methyl paraben (MP) and urinary 2-napthol (2-NAP) were negatively associated with all-cause mortality, while serum cadmium (Cd) was positively associated with all-cause mortality. Urinary bisphenol A (BPA) was positively associated with CVD mortality.

Nuclear Magnetic Resonance Chemical Shift As Highly Explainable Chemical Structure Fingerprints for Anion Exchange Membrane Polymers

Recent shift towards clean energy increased the demand for both fuel cells (used as clean power generator) and water electrolyzer (used as hydrogen supply) significantly. Anion exchange membrane (AEM) serves as a core component for these devices, though its low anion conductivity and durability inhibit their potential for commercialization. Many research and development (R&D) have been done seeking for improvements in AEM1, but current empirical-centric method consumes significant amount of resources, such as cost, labor, and time. To reduce resource consumptions, implementing materials informatics (MI) that allows high-speed screening of materials through a pre-trained AEM polymer machine learning (ML) model is important. However, AEMs are made up of polymers whose chemical structures are complex and hard to represent in a machine-understandable form. Fingerprints are often used to represent chemical structures in numerical forms generated through algorithm2. Majority of these fingerprints are designed for small molecules, not polymers, but they are usually unintuitive and difficult to understand due to their topological nature. In contrast, nuclear magnetic resonance (NMR) chemical shift have long been used in chemistry to identify the chemical structure of a particular sample3. In this study, we aim to utilize the high-resolution nature of NMR chemical shift to identify structural formula as chemical structure fingerprint for ML model, such that a polymer-suited and highly explainable fingerprint can be developed.First, an AEM database containing structural and experimental condition information was built using data extracted from 62 papers. Experimental conditions included were anion conductivity measuring temperature, alkaline stability test measuring condition (test temperature, length of test, and concentration of alkaline solution). Then, the 13C NMR chemical shift for the chemical structure contained in structural information was calculated using ChemDraw. The obtained chemical shifts were converted to numerical strings and is named as “NMR fingerprint”. A new AEM database containing both structural (molar ratio of each building blocks and NMR fingerprints) and experimental condition was used as the training database for ML models. Target variable was set to anion conductivity, and the rest were explanatory variables. ML model used was XGBoost. Cross-validation was used to evaluate the capability of ML models to predict anion conductivity of novel AEM polymers. Prediction logic was analyzed using Shapley additive explanations (SHAP) value.The database built contains data from 62 AEM papers, with 2,197 anion conductivity data points. Each AEM chemical structures present in the database was converted to NMR fingerprints using NMR chemical shifts, obtaining around 2,000 NMR fingerprints for each AEM polymer unit. Together with the experimental conditions and structural information included in the database, the data were used as train-validation dataset for XGBoost. The coefficient of determination (R2) obtained for cross-validated model was 0.9235, implying that the model learnt and determined the relationship between anion conductivity and AEM polymer structure with high accuracy, with the aid of experimental conditions. Then, the prediction logic of the ML model was explored using SHAP values, which are values computed from coalitional game theory, and is used to increase transparency and interpretability of ML models. Analyzing the plot of SHAP values for top 20 important variables used in XGBoost showed that measuring temperature for anion conductivity ranked highest, which is in coherence to the well-known behavior of AEM polymers. Besides, non-experimental condition variables such as 29.8_A ranked into the top 3 important variables. 29.8_A is the chemical shift for alkyl groups attached in between two imidazolium group of AEM polymer, suggesting that the presence or absence of more than one imidazolium group per side chain is important to determine the anion conductivity of an AEM polymer. SHAP values for 29.8_A show that higher feature value (pink color) gives higher impact (positive region of x-axis) to the target variable, inferring that having alkyl groups between imidazolium groups give beneficial effect to anion conductivity. Such ability to explain the prediction logic of ML model shows that using NMR chemical shifts as fingerprints for AEM polymer structures provide intuitive, human-understandable ML prediction logic explanation. Together with the high cross-validation accuracy, NMR chemical shifts hold the potential to not only be a gold standard in expressing polymer structures in machine-understandable form, but also to strongly push the adoption of ML in the AEM polymer field, creating a paradigm shift for AEM R&D.Reference S. Gottesfeld et al., J. Power Sources 2018, 375, 170-184.J. Bajorath, J. Chem. Inf. Comput. Sci. 2001, 41, 2, 233–245.P. Jezzard et al., Adv. Mater. 1992, 4, 2, 82-90. Figure 1

Predicting positron emission tomography brain amyloid positivity using interpretable machine learning models with wearable sensor data and lifestyle factors

BackgroundDeveloping a screening method for identifying individuals at higher risk of elevated brain amyloid burden is important to reduce costs and burden to patients in clinical trials on Alzheimer’s disease or the clinical setting. We developed machine learning models using objectively measured lifestyle factors to predict elevated brain amyloid burden on positron emission tomography.MethodsOur prospective cohort study of non-demented, community-dwelling older adults aged ≥ 65 years was conducted from August 2015 to September 2019 in Usuki, Oita Prefecture, Japan. One hundred and twenty-two individuals with mild cognitive impairment or subjective memory complaints (54 men and 68 women, median age: 75.50 years) wore wearable sensors and completed self-reported questionnaires, cognitive test, and positron emission tomography imaging at baseline. Moreover, 99 individuals in the second year and 61 individuals in the third year were followed up. In total, 282 eligible records with valid wearable sensors, cognitive test results, and amyloid imaging and data on demographic characteristics, living environments, and health behaviors were used in the machine learning models. Amyloid positivity was defined as a standardized uptake value ratio of ≥ 1.4. Models were constructed using kernel support vector machine, Elastic Net, and logistic regression for predicting amyloid positivity. The mean score among 10 times fivefold cross-validation repeats was utilized for evaluation.ResultsIn Elastic Net, the mean area under the receiver operating characteristic curve of the model using objectively measured lifestyle factors alone was 0.70, whereas that of the models using wearable sensors in combination with demographic characteristics and health and life environment questionnaires was 0.79. Moreover, 22 variables were common to all machine learning models.ConclusionOur machine learning models are useful for predicting elevated brain amyloid burden using readily-available and noninvasive variables without the need to visit a hospital.Trial registrationThis prospective study was conducted in accordance with the Declaration of Helsinki and was approved by the local ethics committee of Oita University Hospital (UMIN000017442). A written informed consent was obtained from all participants. This research was performed based on the Strengthening the Reporting of Observational Studies in Epidemiology reporting guideline.

An interpretable machine learning model for trajectory prediction based on nonlinear dynamics mechanism constraints: applications for HVs

An interpretable machine learning model for trajectory prediction based on nonlinear dynamics mechanism constraints: applications for HVs

Interpretable machine learning model for evaluating mechanical properties of concrete made with recycled concrete aggregate

AbstractThe main objective of this paper is to use the data‐driven approach to predict and evaluate the mechanical properties of concrete made with recycled concrete aggregate (RCA), including compressive strength and elastic modulus. Using 358 data samples, including 10 input variables, 10 popular machine learning (ML) algorithms are introduced to select the best ML performance model for predicting RCA concrete's compressive strength and elastic modulus. Gradient Boosting and Categorial Boosting have the best performance in predicting the compressive strength of RCA concrete, with R2 = 0.9112, RMSE = 5.3464 MPa, MAE = 4.0845 MPa, and R2 = 0.9175, RMSE = 5.1520 MPa, MAE = 3.7567 MPa, respectively. Light Gradient Boosting and Categorial Boosting have the best performance in predicting the elastic modulus of RCA concrete, with R2 = 0.8775, RMSE = 2.3560 GPa, MAE = 1.8330 GPa, and R2 = 0.9300, RMSE = 2.3560 MPa, MAE = 1.2589 MPa, respectively. Based on the Shapley Additive Explanation analysis, the influence of main factors on compressive strength and elastic modulus of RCA concrete values has been analyzed qualitatively and quantitatively. RCA replacement level and cement/sand ratio slightly affect compressive strength but have a dominant influence on the elastic modulus of RCA concrete.

Exploring explainable AI: category theory insights into machine learning algorithms

Explainable artificial intelligence (XAI) is a growing field that aims to increase the transparency and interpretability of machine learning (ML) models. The aim of this work is to use the categorical properties of learning algorithms in conjunction with the categorical perspective of the information in the datasets to give a framework for explainability. In order to achieve this, we are going to define the enriched categories, with decorated morphisms, LearnLearn , ParaPara and MNetMNet of learners, parameterized functions, and neural networks over metric spaces respectively. The main idea is to encode information from the dataset via categorical methods, see how it propagates, and lastly, interpret the results thanks again to categorical (metric) information. This means that we can attach numerical (computable) information via enrichment to the structural information of the category. With this, we can translate theoretical information into parameters that are easily understandable. We will make use of different categories of enrichment to keep track of different kinds of information. That is, to see how differences in attributes of the data are modified by the algorithm to result in differences in the output to achieve better separation. In that way, the categorical framework gives us an algorithm to interpret what the learning algorithm is doing. Furthermore, since it is designed with generality in mind, it should be applicable in various different contexts. There are three main properties of category theory that help with the interpretability of ML models: formality, the existence of universal properties, and compositionality. The last property offers a way to combine smaller, simpler models that are easily understood to build larger ones. This is achieved by formally representing the structure of ML algorithms and information contained in the model. Finally, universal properties are a cornerstone of category theory. They help us characterize an object, not by its attributes, but by how it interacts with other objects. Thus, we can formally characterize an algorithm by how it interacts with the data. The main advantage of the framework is that it can unify under the same language different techniques used in XAI. Thus, using the same language and concepts we can describe a myriad of techniques and properties of ML algorithms, streamlining their explanation and making them easier to generalize and extrapolate.

Local post-hoc interpretable machine learning model for prediction of dementia in young adults

&lt;p&gt;Dementia is still the prevailing brain disease with late diagnosis. There is a large increase in dementia disease among young adults. The major reason is over indulgence of young adults on social media resulting in denial of disease and delayed clinical diagnosis. Dementia is preventable and curable if diagnosed at an early stage, however, no attempts are being made to miti gate dementia in young adults. Today artificial intelligence (AI) based advanced technology with real-life consultations in clinical or remote setups are proved beneficial and is used to detect dementia. Most AI-based test is dependent on computer-aided di agnosis (CAD) tools and uses non-invasive imaging technology such as magnetic resonance imaging (MRI) data for disease diagnosis. In this paper, a local post-hoc interpretable machine learning (LPIML) model for prediction of dementia in young adults is proposed. The performance parameters are computed and compared based on accuracy, specificity, precision, F1 score and recall. The proposed work yields 98.87% training accuracy on original images and 99.31% training accuracy on morphologically enhanced images. The performance results are intrinsic and intuitive in learning the prediction results of individual case. The adoption of the proposed work will accelerate the diagnosis process in the era of digital healthcare.&lt;/p&gt;

Application of Interpretable Machine Learning Models Based on Ultrasonic Radiomics for Predicting the Risk of Fibrosis Progression in Diabetic Patients with Nonalcoholic Fatty Liver Disease

Application of Interpretable Machine Learning Models Based on Ultrasonic Radiomics for Predicting the Risk of Fibrosis Progression in Diabetic Patients with Nonalcoholic Fatty Liver Disease

Development of an interpretable machine learning-based intelligent system of exercise prescription for cardio-oncology preventive care: A study protocol.

Cardiovascular disease (CVD) and cancer are the first and second causes of death in over 130 countries across the world. They are also among the top three causes in almost 180 countries worldwide. Cardiovascular complications are often noticed in cancer patients, with nearly 20% exhibiting cardiovascular comorbidities. Physical exercise may be helpful for cancer survivors and people living with cancer (PLWC), as it prevents relapses, CVD, and cardiotoxicity. Therefore, it is beneficial to recommend exercise as part of cardio-oncology preventive care. With the progress of deep learning algorithms and the improvement of big data processing techniques, artificial intelligence (AI) has gradually become popular in the fields of medicine and healthcare. In the context of the shortage of medical resources in China, it is of great significance to adopt AI and machine learning methods for prescription recommendations. This study aims to develop an interpretable machine learning-based intelligent system of exercise prescription for cardio-oncology preventive care, and this paper presents the study protocol. This will be a retrospective machine learning modeling cohort study with interventional methods (i.e., exercise prescription). We will recruit PLWC participants at baseline (from 1 January 2025 to 31 December 2026) and follow up over several years (from 1 January 2027 to 31 December 2028). Specifically, participants will be eligible if they are (1) PLWC in Stage I or cancer survivors from Stage I; (2) aged between 18 and 55 years; (3) interested in physical exercise for rehabilitation; (4) willing to wear smart sensors/watches; (5) assessed by doctors as suitable for exercise interventions. At baseline, clinical exercise physiologist certificated by the joint training program (from 1 January 2023 to 31 December 2024) of American College of Sports Medicine and Chinese Association of Sports Medicine will recommend exercise prescription to each participant. During the follow-up, effective exercise prescription will be determined by assessing the CVD status of the participants. This study aims to develop not only an interpretable machine learning model to recommend exercise prescription but also an intelligent system of exercise prescription for precision cardio-oncology preventive care. This study is approved by Human Experimental Ethics Inspection of Guangzhou Sport University. http://www.chictr.org.cn, identifier ChiCTR2300077887.

Developing an interpretable machine learning model for predicting COVID-19 patients deteriorating prior to intensive care unit admission using laboratory markers

Coronavirus disease (COVID-19) remains a significant global health challenge, prompting a transition from emergency response to comprehensive management strategies. Furthermore, the emergence of new variants of concern, such as BA.2.286, underscores the need for early detection and response to new variants, which continues to be a crucial strategy for mitigating the impact of COVID-19, especially among the vulnerable population. This study aims to anticipate patients requiring intensive care or facing elevated mortality risk throughout their COVID-19 infection while also identifying laboratory predictive markers for early diagnosis of patients. Therefore, haematological, biochemical, and demographic variables were retrospectively evaluated in 8,844 blood samples obtained from 2,935 patients before intensive care unit admission using an interpretable machine learning model. Feature selection techniques were applied using precision-recall measures to address data imbalance and evaluate the suitability of the different variables. The model was trained using stratified cross-validation with k=5 and internally validated, achieving an accuracy of 77.27%, sensitivity of 78.55%, and area under the receiver operating characteristic (AUC) of 0.85; successfully identifying patients at increased risk of severe progression. From a medical perspective, the most important features of the progression or severity of patients with COVID-19 were lactate dehydrogenase, age, red blood cell distribution standard deviation, neutrophils, and platelets, which align with findings from several prior investigations. In light of these insights, diagnostic processes can be significantly expedited through the use of laboratory tests, with a greater focus on key indicators. This strategic approach not only improves diagnostic efficiency but also extends its reach to a broader spectrum of patients. In addition, it allows healthcare professionals to take early preventive measures for those most at risk of adverse outcomes, thereby optimising patient care and prognosis.

Single Cell Profiling Unravels Plasma Cell Gene Expression Signatures Predictive of Response to Daratumumab Based Therapy, and Prognosis in Multiple Myeloma

Background: Multiple Myeloma (MM) is the second most common hematologic malignancy and remains incurable. Daratumumab (dara) is a potent anti-CD38 monoclonal antibody used for MM treatment, but responses are heterogeneous, and resistance is inevitable. There is hence an urgent need for biomarkers to predict the response to dara. We hypothesized that gene expression profiles (GEP) of tumor subpopulations could be used to predict the response to dara-based therapy. Methodology: We used the Enhanced Single-Cell Analysis with Protein Expression (ESCAPE) RNA-Seq platform (Singleron Biotechnologies) to simultaneously profile the gene and protein expression of plasma cells from the bone marrow of 15 MM patients prior to treatment with dara-based regimens. These data were combined with publicly available single-cell data from Cohen et al. 2021, resulting in a total of 32 samples. Response was defined based on the international myeloma working group (IMWG) criteria, considering patients achieving a very good partial response (VGPR) or better as “responders,” while those achieving a partial response (PR) or worse were deemed “non-responders”. The data were split into 23 training samples (15 responders and 8 non-responders) and 9 validation samples (3 responders and 6 non-responders), each set containing a combination of cases from the two sources of single-cell data. An interpretable machine-learning model was then built to predict the response of each individual cell. The median response score for each sample was used to predict the patient-level response. The Shapley Additive Explanations (SHAP) values from the dara response prediction model were then used as a starting point for creating a general prognostic model. The 200 most influential genes from the dara response prediction model were used to create a linear response model based on the TCGA (MMRF-Compass) bulk RNAseq dataset. The bulk RNAseq dataset was split 80:20 into training and test datasets. Results from the test dataset are presented here. Results: We analyzed single-cell data from a total of 13,094 plasma cells from 32 BM samples. These included 17 cases of relapsed MM described by Cohen et al., who were treated with dara, carfilzomib, lenalidomide, and dexamethasone. The patients treated at our centre comprised 7 cases of relapsed MM treated with dara, thalidomide, and dexamethasone, and 8 samples from patients with newly diagnosed MM treated with dara, bortezomib, and dexamethasone on clinical trials. All 9 patients in the test dataset had their responses correctly predicted (Fig 1). A separate cohort of non-dara-treated patient samples was also tested using the same model, and the response prediction was not accurate (AUC=0.67), suggesting that this model has specificity for predicting response to dara and is not a general response classifier (data not shown). Based on the SHAP values from the dara response prediction model, we developed a prognostic model from bulk RNAseq data consisting of four genes. A combination of this four-gene signature plus the international staging system (ISS) score resulted in a superior prediction of overall survival compared to the ISS alone with a hazard ratio of 13.5 compared to 2.5 for the ISS alone in the same dataset (Fig 2). Conclusions: Combining single-cell RNA sequencing with machine-learning methods may have value in predicting response to immunotherapy in MM. We propose that gene signatures derived from single-cell data may augment clinical decision-making for determining treatment allocation. We also show that these signatures may be combined with existing prognostic scores, resulting in improved delineation of risk groups. Furthermore, the applicability of our gene signature to a bulk gene expression dataset enhances its clinical relevance. Our study demonstrates the power of single-cell omics to identify novel predictors of response to therapy and prognosis in MM, which may be translated into clinical use.

Prediction of phytoplankton biomass and identification of key influencing factors using interpretable machine learning models

The water quality of the Middle Route of the South-to-North Water Diversion Project (MRP) of China is related to the health and safety of about 8500w people. In recent years, multiple occurrences of abnormal algal proliferation in water conveyance channels have posed potential risks to the water quality of diverted water. To clarify the growth situation of phytoplankton and evaluate the response relationships between water environmental parameters and phytoplankton in the MRP, in this study, a long-term monitoring dataset was collected from seven monitoring sites from May 2015 to December 2020 to statistically analyse the spatiotemporal characteristics of planktonic algal cell density (ACD). Based on this, four machine learning models, including multiple linear regression (MLR), support vector regression (SVR), random forest (RF), and extreme gradient boosting (XBGoost), were constructed to predict the ACD, and the SHapely Additive exPlanations (SHAP) method was used to interpret the best prediction model for identifying the response relationship between water environmental parameters and ACD. The results showed that (1) the seven monitoring sites could be divided into two significant groups (i.e., Group I and Group II) by hierarchical clustering analysis (HCA), among which the ACD of Group II was higher than that of Group I. (2) the performance of four prediction models using different evaluation metrics indicated that the RF model surpassed in prediction accuracy compared with the other three models in predicting ACD variations in Group I and Group II. (3) SHAP analysis revealed that the key factors affecting ACD variations in Group I were water temperature (WT), water depth (WD), permanganate index (PI), flow rate (Q) and dissolved oxygen (DO), and the ACD were inhibited when WT was below 23 °C, WD exceeded 6.5 m, and Q was between 100 and 250 m3/s. Additionally, WT, PI, and DO were the most important predictors of ACD in Group II, and ACD were inhibited when WT was below 24 °C, PI was lower than 2.4 mg/L and DO was higher than 9 mg/L. This research provides a theoretical basis and reference for water quality management and algal ecological control of water transfer projects with high spatial heterogeneity.

Interpretable machine learning models for predicting 90-day death in patients in the intensive care unit with epilepsy

PurposeThis study aims to develop a machine learning-based model for predicting mortality risk in patients with epilepsy admitted to the intensive care unit (ICU), providing clinicians with an accurate prognostic tool to guide individualized treatment. MethodsWe collected clinical data from clinical databases (MIMIC IV and eICU-CRD) of epilepsy patients 24 h after ICU admission. The clinical characteristics of ICU patients with epilepsy were carefully feature selected and processed. MIMIC IV as the training set and eICU-CRD database as the test set. Six models were developed and validated, and the best LightGBM model was selected by performance comparison and analysed for interpretability. ResultsThe final cohort comprised 429 patients for training and 1217 for testing. The training set exhibited a 90-day mortality rate of 9.32 %, and the test set had an in-hospital 90-day mortality rate of 4.10 %. Utilizing the LightGBM model, we achieved an AUC of 0.956 in the training set. External validation demonstrated promising results with accuracy of 0.898, precision of 0.975, AUC of 0.781, F1 score of 0.945, highlighting the model's potential for guiding clinical decision-making. Significant factors influencing model performance included the severity of illness, as measured by the OASIS score, and clinical parameters like heart rate and body temperature. ConclusionThis study introduces a machine learning-based approach to predict mortality risk in ICU epilepsy patients, offering a valuable tool for clinicians to identify high-risk individuals and devise personalized treatment strategies, thus improving patient prognosis and treatment outcomes.

Interpretable Machine Learning Model For Heart Disease Prediction

In the medical industry, accurately predicting a patient's likelihood of heart disease requires a high-performance model and explaining how the model arrived at its conclusion. To address this, a study has proposed a way to interpret machine learning models using SHAP and LIME. Four models have been created: Vector Machine, Random Forest, XGBoost, and k-Nearest Neighbor. The SVM and XGBoost models exhibit the highest f1-score performance, reaching up to 88%. These models can then be utilized during the interpretation stage with the aid of SHAP and LIME. Based on the SHAP visualization results, it is evident that the predictions made include various significant variables. Meanwhile, LIME explains the classification of each data point. Additionally, it confirms that SHAP and LIME are valuable tools for interpreting models.

Interpretable and intervenable ultrasonography-based machine learning models for pediatric appendicitis

Appendicitis is among the most frequent reasons for pediatric abdominal surgeries. Previous decision support systems for appendicitis have focused on clinical, laboratory, scoring, and computed tomography data and have ignored abdominal ultrasound, despite its noninvasive nature and widespread availability. In this work, we present interpretable machine learning models for predicting the diagnosis, management and severity of suspected appendicitis using ultrasound images. Our approach utilizes concept bottleneck models (CBM) that facilitate interpretation and interaction with high-level concepts understandable to clinicians. Furthermore, we extend CBMs to prediction problems with multiple views and incomplete concept sets. Our models were trained on a dataset comprising 579 pediatric patients with 1709 ultrasound images accompanied by clinical and laboratory data. Results show that our proposed method enables clinicians to utilize a human-understandable and intervenable predictive model without compromising performance or requiring time-consuming image annotation when deployed. For predicting the diagnosis, the extended multiview CBM attained an AUROC of 0.80 and an AUPR of 0.92, performing comparably to similar black-box neural networks trained and tested on the same dataset.

Interpretable prediction of 3-year all-cause mortality in patients with chronic heart failure based on machine learning

BackgroundThe goal of this study was to assess the effectiveness of machine learning models and create an interpretable machine learning model that adequately explained 3-year all-cause mortality in patients with chronic heart failure.MethodsThe data in this paper were selected from patients with chronic heart failure who were hospitalized at the First Affiliated Hospital of Kunming Medical University, from 2017 to 2019 with cardiac function class III-IV. The dataset was explored using six different machine learning models, including logistic regression, naive Bayes, random forest classifier, extreme gradient boost, K-nearest neighbor, and decision tree. Finally, interpretable methods based on machine learning, such as SHAP value, permutation importance, and partial dependence plots, were used to estimate the 3-year all-cause mortality risk and produce individual interpretations of the model's conclusions.ResultIn this paper, random forest was identified as the optimal aools lgorithm for this dataset. We also incorporated relevant machine learning interpretable tand techniques to improve disease prognosis, including permutation importance, PDP plots and SHAP values for analysis. From this study, we can see that the number of hospitalizations, age, glomerular filtration rate, BNP, NYHA cardiac function classification, lymphocyte absolute value, serum albumin, hemoglobin, total cholesterol, pulmonary artery systolic pressure and so on were important for providing an optimal risk assessment and were important predictive factors of chronic heart failure.ConclusionThe machine learning-based cardiovascular risk models could be used to accurately assess and stratify the 3-year risk of all-cause mortality among CHF patients. Machine learning in combination with permutation importance, PDP plots, and the SHAP value could offer a clear explanation of individual risk prediction and give doctors an intuitive knowledge of the functions of important model components.

Development of an interpretable machine learning model for Ki-67 prediction in breast cancer using intratumoral and peritumoral ultrasound radiomics features.

Traditional immunohistochemistry assessment of Ki-67 in breast cancer (BC) via core needle biopsy is invasive, inaccurate, and nonrepeatable. While machine learning (ML) provides a promising alternative, its effectiveness depends on extensive data. Although the current mainstream MRI-centered radiomics offers sufficient data, its unsuitability for repeated examinations, along with limited accessibility and an intratumoral focus, constrain the application of predictive models in evaluating Ki-67 levels. This study aims to explore ultrasound (US) image-based radiomics, incorporating both intra- and peritumoral features, to develop an interpretable ML model for predicting Ki-67 expression in BC patients. A retrospective analysis was conducted on 263 BC patients, divided into training and external validation cohorts. From intratumoral and peritumoral regions of interest (ROIs) in US images, 849 distinctive radiomics features per ROI were derived. These features underwent systematic selection to analyze Ki-67 expression relationships. Four ML models-logistic regression, random forests, support vector machine (SVM), and extreme gradient boosting-were formulated and internally validated to identify the optimal predictive model. External validation was executed to ascertain the robustness of the optimal model, followed by employing Shapley Additive Explanations (SHAP) to reveal the significant features of the model. Among 231 selected BC patients, 67.5% exhibited high Ki-67 expression, with consistency observed across both training and validation cohorts as well as other clinical characteristics. Of the 1698 radiomics features identified, 15 were significantly correlated with Ki-67 expression. The SVM model, utilizing combined ROI, demonstrated the highest accuracy [area under the receiver operating characteristic curve (AUROC): 0.88], making it the most suitable for predicting Ki-67 expression. External validation sustained an AUROC of 0.82, affirming the model's robustness above a 40% threshold. SHAP analysis identified five influential features from intra- and peritumoral ROIs, offering insight into individual prediction. This study emphasized the potential of SVM model using radiomics features from both intra- and peritumoral US images, for predicting elevated Ki-67 levels in BC patients. The model exhibited strong performance in validations, indicating its promise as a noninvasive tool to enable personalized decision-making in BC care.

Enhancement of atmospheric oxidation capacity induced co-pollution of the O3 and PM2.5 in Lanzhou, northwest China

In recent years, the co-pollution of surface ozone (O3) and fine particulate matter (PM2.5) has emerged as a critical concern within specific regions of China's atmospheric environment. This study employed a comprehensive approach by integrating statistical analysis with the interpretable ensemble machine learning model. Delving deeply into the intricate mechanisms behind O3 and PM2.5 co-pollution in Lanzhou city from 2019 to 2022, the research synthesized and analyzed an array of data sources, including ground observations, a multi-parameter lidar system, and meteorological data. Our findings, derived from ground observations to vertical distribution, unequivocally confirm that the enhancement of atmospheric oxidation capacity serves as a critical driver in the genesis of secondary particles, playing a substantial role in the augmented levels of O3 and PM2.5 experienced during the warm season. Moreover, the impact of local weather patterns is indispensable as it precipitates a relatively stable mid-level atmosphere, culminating in elevated surface concentrations of both PM2.5 and O3. Overall, this study emphatically underscores the importance of adopting a comprehensive approach to address these environmental challenges.

Forecasting carbon price trends based on an interpretable light gradient boosting machine and Bayesian optimization

The future carbon price is crucial to relevant companies, investors, and carbon policymakers, and the significance of carbon price prediction research is self-evident. However, existing study usually predicts actual carbon prices, rarely considering price trends and lacking reasonable interpretations for the prediction model. Thus, in this study, an interpretable machine learning model is proposed to predict carbon price trends. It integrates five methods, including the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), two-stage feature selection (TFS), light gradient boosting machine (LightGBM) optimized by Bayesian optimization algorithm (BOA), and SHapley Additive exPlanations (SHAP). The effectiveness of the proposed model is validated with the carbon prices of the Hubei carbon trading market, which has the largest volume among Chinese markets. The experimental results showed that the proposed model outperforms other benchmark models under five evaluation criteria, including AUC, Accuracy, Precision, Recall, and F1 score, on multiple-step predictions. For one-step-ahead prediction, the average hit ratio results are 0.8342, 77.32 %, 77.87 %, 76.83 %, and 76.88 % respectively; for five-step-ahead prediction, the average hit ratio results are 0.7641, 69.25 %, 71.17 %, 71.97 %, and 71.00 % respectively; and for ten-step-ahead prediction, the average hit ratio results are 0.7519, 69.11 %, 73.80 %, 69.61 %, and 71.16 % respectively. The SHAP model interpretation results indicated that the high-frequency intrinsic mode function (IMF) components of the historical carbon price are the most important features for predicting carbon price trends. This study contributes by forecasting both the upward and downward trends of carbon prices through multi-step-ahead forecasting with the LightGBM model and further interpreting the model’s predictions with the SHAP approach. Therefore, the proposed model has excellent forecasting performance with interpretability, which is an effective tool for forecasting carbon price trends.