Individual Input Research Articles

Three-Axis Vibration Isolation of a Full-Scale Magnetorheological Seat Suspension

This study examines the three-axis vibration isolation capabilities of a full-scale magnetorheological (MR) seat suspension system utilizing experimental methods to assess performance under both single-axis and simultaneous three-axis input conditions. To achieve this, a semi-active MR seat damper was designed and manufactured to address excitations in all three axes. The damper effectiveness was tested experimentally for axial and lateral motions, focusing on dynamic stiffness and loss factor using an MTS machine. Prior to creating the full-scale MR seat suspension, a scaled-down version at one-third size was developed to verify the damper’s ability to effectively reduce vibrations in response to practical excitation levels. Additionally, a narrow-band frequency-shaped semi-active control (NFSSC) algorithm was developed to optimize vibration suppression. Ultimately, a full-scale MR seat suspension was assembled and tested with a 50th percentile male dummy, and comprehensive three-axis vibration isolation tests were conducted on a hydraulic multi-axis simulation table (MAST) for both individual inputs over a frequency range up to 200 Hz and for simultaneous multi-directional inputs. The experimental results demonstrated the effectiveness of the full-scale MR seat suspension in reducing seat vibrations.

Predicting freshwater biological quality using macrophytes: A comparison of empirical modelling approaches

Difficulties have hampered bioassessment in southern European rivers due to limited reference data and the unclear impact of multiple interacting stressors on plant communities. Predictive modelling may help overcome this limitation by aggregating different pressures affecting aquatic organisms and showing the most influential factors. We assembled a dataset of 292 Mediterranean sampling locations on perennial rivers and streams (mainland Portugal) with macrophyte and environmental data. We compared models based on multiple linear regression (MLR), boosted regression trees (BRT) and artificial neural networks (ANNs). Secondarily, we investigated the relationship between two macrophyte indices grounded in distinct conceptual premises (the Riparian Vegetation Index — RVI, and the Macrophyte Biological Index for Rivers — IBMR) and a set of environmental variables, including climatic conditions, geographical characteristics, land use, water chemistry and habitat quality of rivers. The quality of models for the IBMR was superior to those for the RVI in all cases, which indicates a better ecological linkage of IBMR with the stressor and abiotic variables. The IBMR using ANN outperformed the BRT models, for which the r-Pearson correlation coefficients were 0.877 and 0.801, and the normalised root mean square errors were 10.0 and 11.3, respectively. Variable importance analysis revealed that longitude and geology, hydrological/climatic conditions, water body size and land use had the highest impact on the IBMR model predictions. Despite the differences in the quality of the models, all showed similar importance to individual input variables, although in a different order. Despite some difficulties in model training for ANNs, our findings suggest that BRT and ANNs can be used to assess ecological quality, and for decision-making on the environmental management of rivers.

An inference method for global sensitivity analysis

Although there is a plethora of methods to estimate sensitivity indices associated with individual inputs, there is much less work on interaction effects of every order, especially when it comes to make inferences about the true underlying values of the indices. To fill this gap, a method that allows one to make such inferences simultaneously from a Monte Carlo sample is given. One advantage of this method is its simplicity: it leverages the fact that Shapley effects and Sobol indices are only linear transformations of total indices, so that standard asymptotic theory suffices to get confidence intervals and to carry out statistical tests. To perform the numerical computations efficiently, Möbius inversion formulas are used, and linked to the fast Möbius transform algorithm. The method is illustrated on two dynamical systems, both with an application in life sciences: a Boolean network modeling a cellular decision-making process involving 12 inputs, and a system of ordinary differential equations modeling some population dynamics involving 10 inputs.

Machine Learning Driven Fluidity and Rheological Properties Prediction of Fresh Cement-Based Materials.

Controlling workability during the design stage of cement-based material mix ratios is a highly time-consuming and labor-intensive task. Applying artificial intelligence (AI) methods to predict and optimize the workability of cement-based materials can significantly enhance the efficiency of mix design. In this study, experimental testing was conducted to create a dataset of 233 samples, including fluidity, dynamic yield stress, and plastic viscosity of cement-based materials. The proportions of cement, fly ash (FA), silica fume (SF), water, superplasticizer (SP), hydroxypropyl methylcellulose (HPMC), and sand were selected as inputs. Machine learning (ML) methods were employed to establish predictive models for these three early workability indicators. To improve prediction capability, optimized hybrid models, such as Particle Swarm Optimization (PSO)-based CatBoost and XGBoost, were adopted. Furthermore, the influence of individual input variables on each workability indicator of the cement-based material was examined using Shapley Additive Explanations (SHAP) and Partial Dependence Plot (PDP) analyses. This study provides a novel reference for achieving rapid and accurate control of cement-based material workability.

Development of quantitative PET/MR imaging for measurements of hepatic portal vein input function: a phantom study

BackgroundAccurate pharmacokinetic modelling in PET necessitates measurements of an input function, which ideally is acquired non-invasively from image data. For hepatic pharmacokinetic modelling two input functions need to be considered, to account for the blood supply from the hepatic artery and portal vein. Image-derived measurements at the portal vein are challenging due to its small size and image artifacts caused by respiratory motion. In this work we seek to demonstrate, using phantom experiments, how a dedicated PET/MR protocol can tackle these challenges and potentially provide input function measurements of the portal vein in a clinical setup.MethodsA custom 3D printed PET/MR phantom was constructed to mimic the liver and portal vein. PET/MR acquisitions were made with emulated respiratory motion. The PET/MR imaging protocol consisted of high-resolution anatomical MR imaging of the portal vein, followed by a PET acquisition in parallel to a dedicated motion-tracking MR sequence. Motion tracking and deformation information were extracted from PET data and subsequently used in PET reconstruction to produce dynamic series of motion-free PET images. Anatomical MR images were used post PET reconstruction for partial volume correction of the input function measurements.ResultsReconstruction of dynamic PET data with motion-compensation provided nearly motion-free series of PET frame data, suitable for image derived input function measurements of the portal vein. After partial volume correction, the individual input function measurements were within a 16.1% error range from the true activity in the portal vein compartment at the time of PET acquisition.ConclusionThe proposed protocol demonstrates clinically feasible PET/MR imaging of the liver for pharmacokinetic studies with accurate quantification of the portal vein input function, including correction for respiratory motion and partial volume effects.

The Segmented Multi‐Source Sediment Routing System on the Hangingwall Dipslope of the Xihu Depression, East China Sea Shelf Basin: Insights From Palaeogeomorphology, U–Pb Ages and Heavy Minerals

ABSTRACTCoeval input systems in rift basins may interact with each other to form a segmented multi‐source sediment routing system. Importantly, its division into proximal zones, where a single source dominates, and interaction zones, where multiple sources mix, enables the interactions between input systems to be characterised. Here, we exploit this conceptual framework to revisit the middle Eocene–early Oligocene hangingwall dipslope of the Xihu depression in the East China Sea Shelf Basin, where extensive 3D seismic data, detrital zircon U–Pb ages and heavy mineral compositions are available. We first combined palaeogeomorphological and sedimentological features with age signatures to distinguish three areas: the northwestern area was identified for its proximity to the Haijiao uplift and invariably high proportions of Palaeoproterozoic ages (41%–54%); the southwestern area adjacent to the Yushan uplift was distinct for enriched Cretaceous‐aged zircons (36%) and the transition area in between was characterised by its remoteness to both uplifts, an embayed geometry and mixed age signatures that are not identical to any individual input. These spatial variations support the segmented framework for the multi‐source system, with the northwestern and southwestern areas representing two palaeo‐input systems and the transition area as their interaction zone. In this context, we then used mixture models to determine spatio‐temporal variations in the mixing proportions of the two palaeo‐input systems. The zircon‐based results indicate that the mixing proportion sustained from the middle to the late Eocene, during which the basin was in the late syn‐rift stage and marine environments. This is corroborated by heavy mineral composition that shows only minor changes. We interpret the roughly sustained mixing proportions as reflecting both the spatially uniform nature of broad subsidence and the strong tidal processes that ‘erased’ the effects of avulsions. In contrast, a clear provenance shift in both zircon ages and heavy minerals occurred from the late Eocene to the early Oligocene, coinciding with a transition to the tectonic inversion stage and a shift towards non‐marine environments. The provenance shift, together with the southward expansion of the axial drainage, likely represents the sedimentary response to the southward decreasing inversion magnitude of the Yuquan Event. In addition, we hypothesize that in the absence of strong tides, avulsions might have controlled the mixing proportion, particularly over short timescales. Ultimately, this study demonstrates the segmented multi‐source framework, if properly incorporated, can provide key insights into dipslope sedimentation.

Priority algorithms with advice for disjoint path allocation problems

We analyze the Disjoint Path Allocation problem (DPA) in the priority framework. Motivated by the problem of traffic regulation in communication networks, DPA consists of allocating edge-disjoint paths in a graph. Like an online algorithm, a priority algorithm receives its input sequentially and must output irrevocable decisions for individual input items before having seen the entire input. However, in contrast to the online setting, a priority algorithm may choose an order on the set of all possible input items and the actual input is then presented according to this order. A priority algorithm is thus a natural model for the intuitively well-understood concept of a greedy algorithm.Mainly motivated by their application for proving lower bounds, we also consider priority algorithms with advice, thus measuring the necessary amount of information about the yet unknown parts of the input.Besides considering the classical variant of the DPA problem on paths and the related problem of Length-Weighted DPA, we mainly focus on DPA on trees. We show asymptotically matching upper and lower bounds on the advice necessary for optimality in LWDPA and generalize the known optimality result for DPA on paths to trees with maximal degree at most 3. On trees with higher maximal degree, we prove matching upper and lower bounds on the approximation ratio in the advice-free priority setting as well as upper and lower bounds on the advice necessary to achieve optimality.

Development and assessment of a machine learning tool for predicting emergency admission in Scotland

Emergency admissions (EA), where a patient requires urgent in-hospital care, are a major challenge for healthcare systems. The development of risk prediction models can partly alleviate this problem by supporting primary care interventions and public health planning. Here, we introduce SPARRAv4, a predictive score for EA risk that will be deployed nationwide in Scotland. SPARRAv4 was derived using supervised and unsupervised machine-learning methods applied to routinely collected electronic health records from approximately 4.8M Scottish residents (2013-18). We demonstrate improvements in discrimination and calibration with respect to previous scores deployed in Scotland, as well as stability over a 3-year timeframe. Our analysis also provides insights about the epidemiology of EA risk in Scotland, by studying predictive performance across different population sub-groups and reasons for admission, as well as by quantifying the effect of individual input features. Finally, we discuss broader challenges including reproducibility and how to safely update risk prediction models that are already deployed at population level.

Performance support team effectiveness in elite sport: a narrative review

ABSTRACT In pursuit of competitive advantage, elite sport organizations are increasingly relying on the support of diverse sport medicine and sport science staff, who are collectively referred to as the performance support team. Whilst it has been suggested that the accumulative input from diverse multiteam systems has the potential to contribute to a resultant whole that is greater than the sum of its parts, team effectiveness is reliant on more than the mere aggregate of diverse experts. The aim of this narrative review was to appraise, summarize, and apply pertinent performance support team literature to a conceptual framework for teamwork and team effectiveness in sport. It specifically explores team effectiveness, with reference to its inputs (i.e. characteristics of individual, team, and environment) and mediators (i.e. team processes and emergent states). This review provides an insight into the individual (i.e. disciplinary knowledge, technical competency, and interpersonal qualities), team (i.e. team composition and leadership), and external (i.e. hierarchical arrangement and environmental factors) inputs that are necessary for team effectiveness, as well as the mediators (i.e. behavioral processes and emergent states) that translate such inputs into desired outcomes.

A Climatology of the Comprehensive Climate Index and Input Values for Türkiye, 1985–2014

Spatial and temporal variations and trends in the Comprehensive Climate Index (CCI) and relevant input values are presented and discussed for fifteen locations across Türkiye. Environmental conditions that are stressful to livestock are captured in the CCI model metrics, which are determined using input of temperature, humidity, wind speed, and solar radiation data. The period from 1985 to 2014 is used to characterize the climatology of CCI across Türkiye. Given its midlatitude location, pronounced seasonality exists in the individual input values and in CCI index values. Differences across Türkiye in CCI are strongly influenced by elevation and moderated for coastal locations. Warming temperature trends and declining levels of humidity combine to create a pattern where CCI has increased at some locations and declined at other places since 1985.

Navigating financial risk: Unlocking the power of modeling and risk quantification in episode-based payment models.

28 Background: The Centers for Medicare and Medicaid Innovation’s (CMMI) Enhancing Oncology Model (EOM) is a voluntary value-based care (VBC) model that requires physician group practices (PGPs) to assume financial risk and performance accountability for episodes of care related to seven common cancers. However, PGPs often lack the means to assess the financial impact and risks associated with these alternative payment models. To address this, we employed Monte Carlo Simulations (MCS) to calculate the likelihood of outcomes and associated risks. This information can assist PGPs in making informed decisions about assuming risk and participating in the EOM. Methods: Using historical episode files from 2019 to 2022 for 23 PGPs in The US Oncology Network, along with the EOM Price Prediction Model provided by CMMI, we calculated the predicted baseline price and savings per episode. We then determined the average savings for each PGP per time-period and utilized the MCS technique to simulate savings and risk, under each EOM risk arrangement. We shared the simulation results with the PGPs to support their decision-making process and participation. Results: Despite having enough episodes per time-period (ranging from 105 to 6100) for the 23 PGPs, the limited number of time-based populations (only 6) based on case mix and patient volume made traditional probability assessments inadequate. To overcome this, we utilized MCS with 5000 simulations per PGP, assuming a normal distribution based on population characteristics. By assessing the savings outcomes of each simulation, we developed probability distributions and confidence intervals to determine the likelihood of generating shared savings, recoupment, or landing in the neutral zone within the EOM. Factors such as episode counts, baseline price, and risk arrangement had the greatest impact on the range of outcomes. We shared the MCS results with all 23 PGPs, highlighting the limitations of the original data and facilitating their decision-making process. Ultimately, 12 PGPs chose to assume risk based on the probability of financial outcomes with the model, and to select the appropriate risk arrangement in the EOM. Conclusions: Compared to the status quo measures of central dispersion, MCS yields a wide range of expected financial outcomes per PGP. MCS is a robust technique for quantifying outcome probabilities in VBC models despite low data volumes but relies on the representativeness and quality of the original data. Conducting sensitivity analysis with MCS to assess the impact of individual assumptions or data inputs on outcome probabilities is an area for further exploration.

Quantifying the contribution of individual inputs used in Zero Budget Natural Farming

AbstractZero Budget Natural Farming (ZBNF) in Andhra Pradesh promotes home‐made, locally sourced, agrochemical‐free inputs and regenerative land management techniques. Inputs consist of seed treatments (bijamrita), microbial inoculum applied either as a liquid foliar spray (liquid jiwamrita) or solid top dressing (solid jiwamrita) to the soil, and mulching (achhadana). However, some farmers do not use all the recommended inputs. There is a lack of evidence on the effects of partial adoption on the resulting yield and on the contributions of individual inputs to the performance of the overall approach. Controlled field experiments were established over two seasons across four agro‐climatic zones. They consisted of five treatments. A Standard ZBNF treatment, which included application of all four ZBNF amendments (bijamirita, solid jiwamrita, liquid jiwamrita and dead mulch). The subsequent four treatments excluded one of the ZBNF inputs (Minus Bijamrita, Minus Soilid Jiwamrita, Minus Liquid Jiwamrita, and Minus Dead Mulch). Exclusion of each ZBNF input individually resulted in a significantly smaller yield than the treatment where all four inputs were used. However, exclusion of solid jiwamrita, liquid jiwamrita and mulching had a larger yield penalty than exclusion of bijamrita. Partial adoption could therfore impact the efficacy of the ZBNF system to deliver sustainable crop yields and satisfy food security. However, further research is needed to examine the effects of input exclusion in the long term, and possible interactions between different ZBNF inputs.

White-Box Machine Learning Models for Accurate Interfacial Tension Prediction in Hydrogen-Brine Mixtures

Abstract The severity of climate change and global warming necessitates the need for a transition from traditional hydrocarbon-based energy sources to renewable energy sources. One intrinsic challenge of renewable energy sources is their intermittent nature, which can be addressed by transforming excess energy into hydrogen and storing it safely for future use. To securely store hydrogen underground, a comprehensive knowledge of the interactions between hydrogen and residing fluids is required. Interfacial tension is an important variable influenced by cushion gases such as CO2 and CH4. This research developed explicit correlations for approximating the interfacial tension of hydrogen-brine mixture using two advanced machine learning techniques: gene expression programming and the group method of data handling. The Interfacial tension of hydrogen-brine mixture was considered as heavily influenced by temperature, pressure, water salinity, and average critical temperature of the gas mixture. The results indicated a higher performance of the group method of data handling-based correlation, showing an average absolute relative error of 4.53%. Subsequently, Pearson, Spearman, and Kendall methods assessed the influence of individual input variables on the correlations’ outputs. Analysis showed that temperature and average critical temperature of the gas mixture had considerable inverse impacts on the estimated interfacial tension values. Finally, the reliability of the gathered databank and the scope of application for the proposed correlations were verified by the leverage approach by illustrating 97.6% of the gathered data within the valid range of the Williams plot.

Assessment of dual time point protocols to produce parametric Ki images in FDG PET/CT: A virtual clinical study.

This simulation study investigated the feasibility of generating Patlak Ki images using a dual time point (DTP-Ki) scan protocol involving two 3-min/bed routine static PET scans and, subsequently, assessed DTP-Ki performance for an optimal DTP scan time frame combination, against conventional Patlak Ki estimated from complete 0-93min dynamic PET data. Six realistic heterogeneous tumors of different characteristic spatiotemporal [18F]FDG uptake distributions for three noise levels commonly found in clinical studies and 20 noise realizations (N=360 samples) were produced by analytic simulations of the XCAT phantom. Subsequently, DTP-Ki images were generated by performing standard linear indirect Patlak analysis with t* -min (Patlakt* = 12) using a scaled population-based input function (sPBIF) model on 66 combinations of early and late 3-min/bed static whole-body PET reconstructed images. All DTP-Ki images were evaluated against respective DTP-Ki images estimated with Patlakt* = 12 and 0-93min individual input functions (iIFs) and against gold standard Ki images estimated with Patlakt* = 12, 0-93min iIFs and tissue time activity curves from all reconstructed WB passes 12-93min post injection. The optimal combination of early and late frames, in terms of attaining the highest correlation between DTP-Ki with sPBIF and gold standard Ki was also determined from a set of 66 different combinations of 2-min early and late frames. Moreover, the performance of DTP-Ki with sPBIF was compared against that of the retention index (RI) in terms of their correlation to the gold standard Ki. Finally, the feasibility and practicality of DTP protocol in the clinic were assessed through the analysis of nine patients. High correlations(>0.9) were observed between DTP-Ki values from sPBIF and those from iIFs for all evaluated DTP protocols while the mean AUC difference between sPBIF and iIFs was less than 10%. The percentage difference of mean values between DTP-Ki from sPBIF and from iIFs was less than 1%. DTP Ki from sPBIF exhibited significantly higher correlation with gold standard Ki, in contrast to RI, across all 66 DTP protocols (p<0.05 using the two-tailed t-test by Williams) with the highest correlation attained for the 50-53-min early + 90-93-min late scan time frames (optimal DTP protocol). Feasibility of generating Patlak Ki [18F] FDG images from an early and a late post injection 3-min/bed routine static scan using a population-based input function model was demonstrated and an optimal DTP scan protocol was determined. The results indicated high correlations between DTP-Ki and gold-standard Ki images that are significantly larger than those between RI and gold-standard Ki.

An interpretable deep learning framework identifies proteomic drivers of Alzheimer's disease.

Alzheimer's disease (AD) is the leading neurodegenerative pathology in aged individuals, but many questions remain on its pathogenesis, and a cure is still not available. Recent research efforts have generated measurements of multiple omics in individuals that were healthy or diagnosed with AD. Although machine learning approaches are well-suited to handle the complexity of omics data, the models typically lack interpretability. Additionally, while the genetic landscape of AD is somewhat more established, the proteomic landscape of the diseased brain is less well-understood. Here, we establish a deep learning method that takes advantage of an ensemble of autoencoders (AEs) - EnsembleOmicsAE-to reduce the complexity of proteomics data into a reduced space containing a small number of latent features. We combine brain proteomic data from 559 individuals across three AD cohorts and demonstrate that the ensemble autoencoder models generate stable latent features which are well-suited for downstream biological interpretation. We present an algorithm to calculate feature importance scores based on the iterative scrambling of individual input features (i.e., proteins) and show that the algorithm identifies signaling modules (AE signaling modules) that are significantly enriched in protein-protein interactions. The molecular drivers of AD identified within the AE signaling modules derived with EnsembleOmicsAE were missed by linear methods, including integrin signaling and cell adhesion. Finally, we characterize the relationship between the AE signaling modules and the age of death of the patients and identify a differential regulation of vimentin and MAPK signaling in younger compared with older AD patients.

CMAA–AHP: combinatorial multicriteria acceptability analysis with the analytic hierarchy process

AbstractCombinatorial multi-criteria acceptability analysis (CMAA) is a framework for supporting multicriteria group decisions that provides both a detailed analysis of the effects of individual decision-maker inputs as well as interactive guidance for a consensus-building process. The analytical hierarchy process (AHP) is a widely-used model of decision-maker evaluations that is based on pairwise comparisons. The goal of this work is to show how CMAA can be integrated with AHP in order to make its benefits available to AHP users. We use a minimal input format for AHP which avoids a problem with inconsistency and also reduces the cognitive load on the decision-makers. We extend the CMAA method by introducing new judgement and preference sensitivity variables, which provide helpful insights for the facilitator of the group decision. An example illustrates the combined CMAA–AHP method and its ability to deliver consensus in a very small number of iterations. Monte Carlo simulation is used to study the convergence behavior of the method for a range of problem dimensions. It was found that the mean number of steps to reach consensus grows linearly with the number of alternatives and criteria. We consider two previously published group decisions that use the standard AHP approach of averaging decision-maker judgements and preferences. In both cases, CMAA–AHP delivers the same rankings based on the original input. However, the new method also provides insight into each decision and would have been able to guide each group to consensus within a small number of resolution steps.

Robust Input Shaping Commands with First-Order Actuators.

This paper presents robust input shaping commands with first-order actuators utilizing a classical robust input shaper for practical applications in input shaping technology. An ideal input shaping command can deviate due to actuator dynamics so that the modified command has a detrimental effect on the performance of oscillation reduction in feedforward control applications. A zero-vibration-derivative (ZVDF) shaper with first-order actuators is analytically proposed using a phasor-vector approach, an exponential function for the approximation of the dynamic response of first-order actuators and the usage of the ZVD shaper. In addition, an equivalent transformation is utilized based on the superposition principle for the convenient inclusion of first-order actuator dynamics and is applied to the individual segment input command. The residual deflection and robustness of the proposed robust input shaping commands are numerically evaluated and compared with those of a conventional ZVD shaper with respect to the parameter uncertainties of flexible systems and actuators. The robust input shaping commands that are possible with first-order actuators are experimentally validated, presenting a better robustness and residual deflection reduction performance than the classical ZVD shaper on a mini bridge crane.

Short-Term Forecasts of Energy Generation in a Solar Power Plant Using Various Machine Learning Models, along with Ensemble and Hybrid Methods

High-quality short-term forecasts of electrical energy generation in solar power plants are crucial in the dynamically developing sector of renewable power generation. This article addresses the issue of selecting appropriate (preferred) methods for forecasting energy generation from a solar power plant within a 15 min time horizon. The effectiveness of various machine learning methods was verified. Additionally, the effectiveness of proprietary ensemble and hybrid methods was proposed and examined. The research also aimed to determine the appropriate sets of input variables for the predictive models. To enhance the performance of the predictive models, proprietary additional input variables (feature engineering) were constructed. The significance of individual input variables was examined depending on the predictive model used. This article concludes with findings and recommendations regarding the preferred predictive methods.

Reliability estimation for individual predictions in machine learning systems: A model reliability-based approach

The conventional aggregated performance measure (i.e., mean squared error) with respect to the whole dataset would not provide desired safety and quality assurance for each individual prediction made by a machine learning model in risk-sensitive regression problems. In this paper, we propose an informative indicator ℛx to quantify model reliability for individual prediction (MRIP) for the purpose of safeguarding the usage of machine learning (ML) models in mission-critical applications. Specifically, we define the reliability of a ML model with respect to its prediction on each individual input x as the probability of the observed difference between the prediction of ML model and the actual observation falling within a small interval when the input x varies within a small range subject to a preset distance constraint, namely ℛx=Py∗−ŷ∗≤εx∗∈Bx, where y∗ denotes the observed target value for the input x∗,ŷ∗ denotes the model prediction for the input x∗, and x∗ is an input in the neighborhood of x subject to the constraint Bx=x∗x∗−x≤δ. The developed MRIP indicator ℛx provides a direct, objective, quantitative, and general-purpose measure of “reliability” or the probability of success of the ML model for each individual prediction by fully exploiting the local information associated with the input x and ML model. Next, to mitigate the intensive computational effort involved in MRIP estimation, we develop a two-stage ML-based framework to directly learn the relationship between x and its MRIP ℛx, thus enabling to provide the reliability estimate ℛx for any unseen input instantly. Thirdly, we propose an information gain-based approach to help determine a threshold value pertaing to ℛx in support of decision makings on when to accept or abstain from counting on the ML model prediction. Comprehensive computational experiments and quantitative comparisons with existing methods on a broad range of real-world datasets reveal that the developed ML-based framework for MRIP estimation shows a robust performance in improving the reliability estimate of individual prediction, and the MRIP indicator ℛx thus provides an essential layer of safety net when adopting ML models in risk-sensitive environments.

Compressive strength prediction of cement base under sulfate attack by machine learning approach

Cement-based materials in underground structures, bridge foundation, and offshore platforms often face sulfate environments. It has been widely reported that the compressive strength (CS) of these materials post sulfate attack is crucial for structural stability and durability. However, existing experimental and analytical methods require extensive resources and lengthy period. For addressing this challenge, therefore, a novel stacking model is proposed to accurately predict the CS of cement base under sulfate attack in this study. The stacking model integrates Support Vector Regression (SVR), Random Forest (RF), and Extreme Gradient Boosting (XGB) as base learners, with Linear Regression (LR) as the meta-learner. In this novel model training, the 330 data sets used are obtained from the relevant literatures. The water-cement ratio, cement content, water content, sand content, sulfate concentration, wetting temperature, wetting time, drying temperature, drying time, exposure time are selected as inputs while the output is CS of cement-based materials subjected to sulphate deterioration. To improve the forecast capability, particle swarm optimization (PSO) is adopted for optimizing hyperparameters of stacking model, which can be implemented by reducing the discrepancy between model prediction and measured CS. By comparing with the results of experiments and the solely machine learning (ML) models, it is found that the stacking model optimized by PSO presents the best prediction accuracy of R²=0.992, RMSE=2.248 MPa, and MAE=1.782 MPa. Moreover, the examination of the influence of individual input variables on the CS of cement-based material subjected to sulfate attack is conducted through the application of Shapley Additive Explanations (SHAP) and Partial Dependence Plot (PDP) analyses. The PSO optimized stacking model proposed in this study can effectively predict the CS of cement-based materials under sulfate attack. Our study provides a new reference for future research in this field.