Relative Importance Of Variables Research Articles

Estimating urban PM2.5 concentration: An analysis on the nonlinear effects of explanatory variables based on gradient boosted regression tree

It is important for air pollution study and control to understand the relationships between the concentrations of fine particulate matter and the influencing factors. This study builds a PM2.5 concentration estimation model with high accuracy considering major factors including meteorological conditions, traffic conditions, and urban spatial layouts based on Gradient Boosted Regression Tree approach. Based on the proposed model, the relative importance of the explanatory variables is presented. The meteorological variables, traffic condition variables and urban spatial layout variables have the predictive powers of 64.6%, 21.2% and 14.2% respectively. The nonlinear relationships between each variable and urban PM2.5 concentration are visualized by partial dependency plots and discussed in detail. The results show that in order not to exceed the limitation of regional PM2.5 concentration, the local government can lower city's temperature, improve urban ventilation and precipitation, alleviate traffic congestions, balance the road construction and PM2.5 generation, increase urban green area and concentrate Points of Interests.

Application of machine learning algorithms in predicting the photocatalytic degradation of perfluorooctanoic acid

ABSTRACT Perfluorooctanoic acid (PFOA) is used in a variety of industries and is highly persistent in the environment, with potential human health risks. Photocatalysis has been extensively used for the decomposition of various organic pollutants, yet its simulation and modeling are challenging. This research aimed to establish different machine learning (ML) algorithms which can simulate and predict the photocatalytic degradation of PFOA. The published results were used to estimate and predict the photocatalytic degradation of PFOA. Statistical criteria including the coefficient of determination (R2 ), mean absolute error (MAE), and mean squared error (MSE) were considered in assessing the best method of modeling. Among the seven ML algorithms pre-screened, Adaptive Boosting (AdaBoost), Gradient Boosting Machine (GBM), and Random Forest (RF) showed the best performance and were chosen for deep modeling and analysis. Grid search was used to optimize the models developed by AdaBoost, GBM, and RF; and permutation variable importance (PVI) was used to analyze the relative importance of different variables. Based on the modeling results, GBM model (R2 = 0.878, MSE = 106.660, MAE = 6.009) and RF model (R2 = 0.867, MSE = 107.500, MAE = 6.796) showed superior performances compared with AdaBoost model (R2 = 0.574, MSE = 388.369, MAE = 16.480). Furthermore, the PVI results suggested that the GBM model provided the best outcome, with the light irradiation time, type of catalyst, dosage of catalyst, solution pH, irradiation intensity, initial PFOA concentration, oxidizing agents (peroxymonosulfate, ammonium persulfate, and sodium persulfate), irradiation wavelength, and solution temperature as the most important process variables in decreasing order.

Differences of education level and job position in digital transformation adoption in Vietnam’s creative industries

The aim of this paper is to contribute to understanding the differences in educational level and job titles that affect the adoption of digital transformation in the context of creative industries. The approach is based primarily on the technology-organizational-environment framework, one-way analysis of variance, and multivariate linear regression that were used to test the empirical model. This article describes the difference between the educational level and the job position groups for the adoption of digital transformation in Vietnam’s creative industries, as well as an essential difference between them for the predictor variables. Our research has underlined the importance of demographic variables in relation to the adoption of digital transformation. The current study is limited to some fields of creative industries in Vietnam, such as design, creative services, new media, audiovisuals, publishing and printed media, and selected demographic parameters. More research should be done on certain fields in the creative industries and other demographic characteristics.

A hybrid approach to forecasting futures prices with simultaneous consideration of optimality in ensemble feature selection and advanced artificial intelligence

The paper presents a framework to forecast futures prices of stocks listed on the National Stock Exchange (NSE) in India during normal (unaffected by the COVID-19 pandemic) and new normal times (affected by COVID-19 and a macroeconomic slowdown). The model leverages a structural model that determines the relevance of the explanatory features used in the study; namely, spot prices, market sentiment, sectoral outlook, historic and implied volatility, crude price volatility, and exchange rate volatility. The proposed Ensemble Feature Selection (EFS) methodology comprising Boruta and Regularized Random Forest (RRF) algorithms is used to screen the explanatory features. Two advanced Artificial Intelligence techniques—Regularized Greedy Forest (RGF) and Deep Neural Network (DNN)—are used in conjunction with Kernel Principal Component Analysis (KPCA) and Autoencoder (AE) for forecasting. To understand the extent and nature of the influence of the explanatory variables, the Explainable Artificial Intelligence (AI) approach has been used. Statistical checks confirm that our hybrid framework is effective. The results indicate that the relative importance of the explanatory variables in forecasting futures prices differs depending on the company concerned and the period under consideration.

Channel Conveyance Variability can Influence Flood Risk as Much as Streamflow Variability in Western Washington State

AbstractChanges in the severity and likelihood of flooding events are typically associated with changes in the intensity and frequency of streamflows, but temporal adjustments in a river's conveyance capacity can also contribute to shifts in flood hazard. To assess the relative importance of channel conveyance to flood hazard, we compare variations in channel conveyance to variations in the flow magnitude of moderate (1.2 years) floods at 50 river gauges in western Washington State between 1930 and 2020. In unregulated rivers, moderate floods have increased across the region, but in regulated rivers this trend is suppressed and in some cases reversed. Variations in channel conveyance are ubiquitous, but the magnitude and timing of adjustments are not regionally uniform. At 40% of gages, conveyance changes steadily and gradually. More often, however, conveyance variability is nonlinear, consisting of multidecadal oscillations (36% of gages), rapid changes due to unusually large sediment‐supply events (14% of gages), and increases or decreases to conveyance following flow regulation (10% of gages). The relative importance of conveyance variability for flood risk depends on the mode of adjustment; in certain locations with historic landslides, extreme floods, and flow regulation, the influence of conveyance changes on flood risk matches or exceeds that of streamflow at the same site. Flood hazard management would benefit from incorporating historic long‐term and short‐term conveyance changes in predictions of future flood hazard variability.

Predicting nitrate leaching loss in temperate rainfed cereal crops: relative importance of management and environmental drivers

Nitrate (NO3) leaching from agriculture represents the primary source of groundwater contamination and freshwater ecosystem degradation. At the field level, NO3 leaching is highly variable due to interactions among soil, weather and crop management factors, but the relative effects of these drivers have not been quantified on a global scale. Using a global database of 82 field studies in temperate rainfed cereal crops with 961 observations, our objectives were to (a) quantify the relative importance of environmental and management variables to identify key leverage points for NO3 mitigation and (b) determine associated changes in crop productivity and potential tradeoffs for high and low NO3 loss scenarios. Machine learning algorithms (XGboost) and feature importance analysis showed that the amount and intensity of rainfall explained the most variability in NO3 leaching (up to 24 kg N ha−1), followed by nitrogen (N) fertilizer rate and crop N removal. In contrast, other soil and management variables such as soil texture, crop type, tillage and N source, timing and placement had less importance. To reduce N losses from global agriculture under changing weather and climatic conditions, these results highlight the need for better targeting and increased adoption of science-based, locally adapted management practices for improving N use efficiency. Future policy discussions should support this transition through different instruments while also promoting more advanced weather prediction analytics, especially in areas susceptible to extreme climatic variation.

Climate change reshuffles northern species within their niches

Climate change is a pervasive threat to biodiversity. While range shifts are a known consequence of climate warming contributing to regional community change, less is known about how species’ positions shift within their climatic niches. Furthermore, whether the relative importance of different climatic variables prompting such shifts varies with changing climate remains unclear. Here we analysed four decades of data for 1,478 species of birds, mammals, butterflies, moths, plants and phytoplankton along a 1,200 km high latitudinal gradient. The relative importance of climatic drivers varied non-uniformly with progressing climate change. While species turnover among decades was limited, the relative position of species within their climatic niche shifted substantially. A greater proportion of species responded to climatic change at higher latitudes, where changes were stronger. These diverging climate imprints restructure a full biome, making it difficult to generalize biodiversity responses and raising concerns about ecosystem integrity in the face of accelerating climate change.

Long-term survival in newly-diagnosed ovarian cancer patients.

e17543 Background: A proportion of women with ovarian cancer, even those with advanced-stage disease, will experience long-term survival post diagnosis. Mixture cure modeling provides insight into factors related to both short-term recurrence and death as well as long-term survival. The goals of this study were to estimate the long-term survival rate, and to evaluate associations between patient characteristics and the long-term survival rate and landmark 5- and 10- year survival, in a sample of newly-diagnosed ovarian cancer patients. Methods: The Tempus EMR dataset contains patients from academic & community oncology centers in the U.S. This study included adult women with a new primary diagnosis of ovarian, fallopian tube, or peritoneal cancer between 2000 and 2020. Women treated with a poly-ADP ribose polymerase inhibitor (PARPi), pembrolizumab, or nivolumab were excluded (n = 947; final sample size = 6,598). Mixture cure models were used to estimate the long-term survival rate (probability of living as long as an age- and gender-matched person without ovarian cancer), overall and stratified by patient demographic & clinical characteristics. Relative importance of variables associated with long-term survival was evaluated using logistic regression models with 5- and 10-year landmark outcomes. Results: Younger age, earlier stage, lower grade, favorable ECOG score, and platinum sensitivity were associated with better overall survival (OS). The long-term survival rate in our model was 25.0% (95% CI: 21.5-28.9). The long-term survival rate was higher for age < 65 vs. age 75+ [29.0% (24.8-33.6) vs. 10.5% (2.9-31.6)], early vs. advanced stage disease [62.0% (41.4-78.9) vs. 17.8% (14.1-22.2)], grade 1-2 vs. 3-4 [31.4% (12.1-60.3) vs. 20.4% (14.6-27.8)], ECOG 0-2 vs. 3-4 [26.1% (21.4-31.3) vs. 2.6% (0.05- 56.5)], and platinum sensitive vs. platinum resistant patients [47.5% (43.1-51.9) vs. 9.4% (7.0-12.5)]. Analyzing 5- and 10-year landmarks [n = 3,944 (59.8%), n = 3,107 (47.1%), respectively] in multivariable logistic regression models, platinum sensitivity had the strongest association, followed by cancer stage. Conclusions: Understanding differences in long-term survival based on patient characteristics is important for clinical decision-making and personalizing treatment. Overall, one quarter of ovarian cancer patients were long-term survivors; long-term survivors were more likely to be younger and have earlier stage disease, low grade disease, better performance status, or sensitivity to platinum-based therapy. These results also provide a baseline to evaluate long-term survival in patients on newer therapies such as PARPi when sufficient follow-up data are available.

A continuous times-series and discrete measure analysis of two individual divers performing the 3½ pike somersault dive

ABSTRACT Springboard diving training is often focused upon skill repetition to establish movement accuracy, stability and consistency. Within-participant study designs provide the ability to understand how individuals create these skills under different movement strategies. IMUs measured angular velocity time-series data of two athletes performing multiple repetitions of forward 3½ somersault pike dives. Functional Principal Component Analyses (fPCA) were performed to examine individual movement structure and variability. The first five fPC’s represented approximately 98% of the variability in angular velocity for both divers. To determine the relative importance of angular velocity variability, Pearson’s correlations for pairwise comparisons were used to assess the relationship between fPC scores and discrete performance variables during takeoff, flight and entry. Divers exhibited a different number and types of significant correlations (International = 4; National = 11). Only one correlation was common for both divers; higher angular velocity during Initial Flight and/or Somersault phases resulted in more vertically aligned entry posture (International: fPC1 r = −0.761, p < 0.05; National: fPC3 r = −0.796, p < 0.01). Findings identify individualised angular velocity time-series structure and kinematic performance variables (International = angular; National = linear) that can be used by coaching/sport science teams to optimisation performance success.

Compressive Strength Prediction of Fly Ash Concrete Using Machine Learning Techniques

It is time-consuming and uneconomical to estimate the strength properties of fly ash concrete using conventional compression experiments. For this reason, four machine learning models—extreme learning machine, random forest, original support vector regression (SVR), and the SVR model optimized by a grid search algorithm—were proposed to predict the compressive strength of fly ash concrete on 270 group datasets. The prediction results of the proposed model were compared using five evaluation indices, and the relative importance and effect of each input variable on the output compressive strength were analyzed. The results showed that the optimized hybrid model showed the best predictive behavior compared to the other three models, and can be used to forecast the compressive strength of fly ash concrete at a specific mix design ratio before conducting laboratory compression tests, which will save costs on the specimens and laboratory tests. Among the eight input variables listed, age and water were the two relatively most important features with superplasticizer and fly ash being of weaker relative importance.

Machine Learning and Regression Analysis Reveal Different Patterns of Influence on Net Ecosystem Exchange at Two Conifer Woodland Sites

Variations in net ecosystem exchange (NEE) of carbon dioxide, and the variables influencing it, at woodland sites over multiple years determine the long term performance of those sites as carbon sinks. In this study, weekly-averaged data from two AmeriFlux sites in North America of evergreen woodland, in different climatic zones and with distinct tree and understory species, are evaluated using four multi-linear regression (MLR) and seven machine learning (ML) models. The site data extend over multiple years and conform to the FLUXNET2015 pre-processing pipeline. Twenty influencing variables are considered for site CA-LP1 and sixteen for site US-Mpj. Rigorous k-fold cross validation analysis verifies that all eleven models assessed generate reproducible NEE predictions to varying degrees of accuracy. At both sites, the best performing ML models (support vector regression (SVR), extreme gradient boosting (XGB) and multi-layer perceptron (MLP)) substantially outperform the MLR models in terms of their NEE prediction performance. The ML models also generate predicted versus measured NEE distributions that approximate cross-plot trends passing through the origin, confirming that they more realistically capture the actual NEE trend. MLR and ML models assign some level of importance to all influential variables measured but their degree of influence varies between the two sites. For the best performing SVR models, at site CA-LP1, variables air temperature, shortwave radiation outgoing, net radiation, longwave radiation outgoing, shortwave radiation incoming and vapor pressure deficit have the most influence on NEE predictions. At site US-Mpj, variables vapor pressure deficit, shortwave radiation incoming, longwave radiation incoming, air temperature, photosynthetic photon flux density incoming, shortwave radiation outgoing and precipitation exert the most influence on the model solutions. Sensible heat exerts very low influence at both sites. The methodology applied successfully determines the relative importance of influential variables in determining weekly NEE trends at both conifer woodland sites studied.

Prediction models for early diagnosis of actinomycotic osteomyelitis of the jaw using machine learning techniques: a preliminary study.

BackgroundThis study aimed to develop and validate five machine learning models designed to predict actinomycotic osteomyelitis of the jaw. Furthermore, this study determined the relative importance of the predictive variables for actinomycotic osteomyelitis of the jaw, which are crucial for clinical decision-making.MethodsA total of 222 patients with osteomyelitis of the jaw were analyzed, and Actinomyces were identified in 70 cases (31.5%). Logistic regression, random forest, support vector machine, artificial neural network, and extreme gradient boosting machine learning methods were used to train the models. The models were subsequently validated using testing datasets. These models were compared with each other and also with single predictors, such as age, using area under the receiver operating characteristic (ROC) curve (AUC).ResultsThe AUC of the machine learning models ranged from 0.81 to 0.88. The performance of the machine learning models, such as random forest, support vector machine and extreme gradient boosting was significantly superior to that of single predictors. Presumed causes, antiresorptive agents, age, malignancy, hypertension, and rheumatoid arthritis were the six features that were identified as relevant predictors.ConclusionsThis prediction model would improve the overall patient care by enhancing prognosis counseling and informing treatment decisions for high-risk groups of actinomycotic osteomyelitis of the jaw.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12903-022-02201-6.

Artificial neural network model for prediction of viscoelastic behaviour of polycarbonate composites

Polymer composites are the result of incorporation of nanoparticles into the polymers and can lead to improvements even with a very small amount of reinforcement which can be tuned according to the applications. In order to understand the behaviour of these polymer composites we need to perform a number of characterizations and analyses which in turn requires investment of money and time. Thus, to reduce the number of characterizations and analyses for developing polymer composites, computational techniques can prove helpful. By means of a computational technique known as artificial neural network (ANN), prediction of the thermo-mechanical properties was made possible. Here dynamic mechanical analysis (DMA) data set was used for characterization of polycarbonate / calcium carbonate-SiO2 core shell composites (polycarbonate composites). Using the dataset, the selected ANN model consisted of a network of [3-10-1]. The prediction accuracy achieved using ANN method, was around 90%. Applicability and performance of ANN to the existing system was also confirmed by mean squared error (MSE), which is favourably small for this case, in the range of 10-5. The output predicted by ANN had a coefficient of correlation of 0.999. Furthermore, sensitivity analysis confirmed the importance of various input variables in relation with output. An optimization of the variables facilitated to maximize the conditions thus predicting glass transition temperature.

Pteridophyte species richness in the central Himalaya is limited by cold climate extremes at high elevations and rainfall seasonality at low elevations.

There is a consensus that climate factors strongly influence species richness along elevation gradients, but which factors are crucial and how they operate are still elusive. Here, we assess the relative importance of temperature‐related versus precipitation‐related variables and the relative importance of extreme climate versus climate seasonality in driving pteridophyte species richness. We used correlation and regression analyses to relate species richness of pteridophytes, and their two major groups (lycophytes, ferns), in fifty 100‐m vertical bands to climatic factors representing different aspects of climatic conditions (general climate, stressful climate, and climate seasonality). Variation partitioning analysis was used to determine the relative importance of each group of climatic factors on species richness. Across the entire elevational gradient, species richness had a parabolic response to mean annual temperature (adjusted R 2 = .87−.91), and a linear response to annual precipitation (adjusted R 2 = .82). Mean annual temperature and annual precipitation in the second‐order polynomial model together explained 96.3%−98.7% of the variation in species richness. The variation in species richness uniquely explained by minimum temperature of the coldest month was much greater than that uniquely explained by temperature seasonality, but the variation in species richness uniquely explained by precipitation during the driest month was much smaller than that uniquely explained by precipitation seasonality. Overall, extreme climate variables explained slightly more variation than did climate seasonality. Our study suggests that pteridophyte richness along the elevational gradient is largely driven by a combination of both temperature‐ and precipitation‐related parameters, although precipitation‐related variables play a slightly stronger role, and that extreme low temperature events (at high elevations) and seasonal precipitation variability (at low elevations) are the strongest determinants of pteridophyte species richness.

Evaluation of Heterocyclic Aromatic Compound Dye (Methylene Blue) on Chitosan Adsorbent Sourced from African Snail Shell: Modelling and Optimization Studies

In this article, the modelling and optimization of five operational process parameters involving initial concentration, adsorbent dosage, contact time, temperature and pH of the solution as it affects the treatment of aqueous solution contaminated with methylene blue, a heterocyclic aromatic compound, on chitosan sourced from African Snail Shell were studied using response surface methodology (RSM) and artificial neural network (ANN) techniques coupled with genetic algorithm. The single and interactive effects of the variables were examined by way of analysis of variance (ANOVA). A comparison of the model techniques was done and an evaluation was carried out with some selected error functions. Both modelling and optimization tools performed creditably well. However, the hybrid ANN-GA proved to be a superior modelling and optimization technique with excellent generalization ability which gave an average absolute deviation between the experimental and predicted data of both response variables considered. The insightful relative importance of the process variables based on the renowned Garson and Olden’s algorithm methods coupled with step by step approach initiated in the Matlab environment were equally investigated. The findings from this study revealed in clear terms that pH and initial concentrations were the most influential parameters and the maximum value of 99.28% of methylene blue removed at optimum conditions affirmed that the chitosan adsorbent is viable for the treatment of effluents from the textile industry.&#x0D;

Dynamics of the stream–lake transitional zone affect littoral lake metabolism

Lake ecosystems, as integrators of watershed and climate stressors, are sentinels of change. However, there is an inherent time-lag between stressors and whole-lake response. Aquatic metabolism, including gross primary production (GPP) and respiration (R), of stream–lake transitional zones may bridge the time-lag of lake response to allochthonous inputs. In this study, we used high-frequency dissolved oxygen data and inverse modeling to estimate daily rates of summer epilimnetic GPP and R in a nutrient-limited oligotrophic lake at two littoral sites located near different major inflows and at a pelagic site. We examined the relative importance of stream variables in comparison to meteorological and in-lake predictors of GPP and R. One of the inflow streams was substantially warmer than the other and primarily entered the lake’s epilimnion, whereas the colder stream primarily mixed into the metalimnion or hypolimnion. Maximum GPP and R rates were 0.2–2.5 mg O2 L−1 day−1 (9–670%) higher at littoral sites than the pelagic site. Ensemble machine learning analyses revealed that > 30% of variability in daily littoral zone GPP and R was attributable to stream depth and stream–lake transitional zone mixing metrics. The warm-stream inflow likely stimulated littoral GPP and R, while the cold-stream inflow only stimulated littoral zone GPP and R when mixing with the epilimnion. The higher GPP and R observed near inflows in our study may provide a sentinel-of-the-sentinel signal, bridging the time-lag between stream inputs and in-lake processing, enabling an earlier indication of whole-lake response to upstream stressors.

Artificial Neural Network-Forecasted Compression Strength of Alkaline-Activated Slag Concretes

The utilization of ordinary Portland cement (OPC) in conventional concretes is synonymous with high carbon emissions. To remedy this, an environmentally friendly concrete, alkaline-activated slag concrete (AASC), where OPC is completely replaced by ground granulated blast-furnace slag (GGBFS) industrial waste, is one of the currently pursued research interests. AASC is not commonly used in the construction industry due to limitations in experience and knowledge on the mix proportions and mechanical properties. To circumvent great labour in the experimental works toward the determination of the optimal properties, this study, therefore, presents the compressive strength prediction of AASC by employing the back-propagation artificial neural network (ANN) modelling technique. To construct this model, a sufficiently equipped experimental databank was built from the literature covering varied mix proportion effects on the compressive strength of AASC. For this, four model variants with different input parameter considerations were examined and the ideal ANN architecture for each model with the best input number–hidden layer neuron number–output number format was identified to improve its prediction accuracy. From such a setting, the most accurate prediction model with the highest determination coefficient, R2, of 0.9817 was determined, with an ANN architecture of 8-18-1 containing inputs such as GGBFS, a fine to total aggregate ratio, sodium silicate, sodium hydroxide, mixing water, silica modulus of activator, percentage of sodium oxide and water–binder ratio. The prediction accuracy of the optimal ANN model was then compared to existing ANN-based models, while the variable selection was compared to existing AASC models with other machine learning algorithms, due to limitations in the ANN-based model. To identify the parametric influence, the individual relative importance of each input variable was determined through a sensitivity analysis using the connection weight approach, whose results indicated that the silica modulus of the activator and sodium silicate greatly affected the AASC compressive strength. The proposed methodology demonstrates that the ANN-based model can predict the AASC compressive strength with a high accuracy and, consequently, aids in promoting the utilization of AASC in the construction industry as green concrete without performing destructive tests. This prediction model can also accelerate the use of AASC without using a cement binder in the concrete matrix, leading to produce a sustainable construction material.

Enabling Knowledge Discovery in Multi-Objective Optimizations of Worker Well-Being and Productivity

Usually, optimizing productivity and optimizing worker well-being are separate tasks performed by engineers with different roles and goals using different tools. This results in a silo effect which can lead to a slow development process and suboptimal solutions, with one of the objectives, either productivity or worker well-being, being given precedence. Moreover, studies often focus on finding the best solutions for a particular use case, and once solutions have been identified and one has been implemented, the engineers move on to analyzing the next use case. However, the knowledge obtained from previous use cases could be used to find rules of thumb for similar use cases without needing to perform new optimizations. In this study, we employed the use of data mining methods to obtain knowledge from a real-world optimization dataset of multi-objective optimizations of worker well-being and productivity with the aim to identify actionable insights for the current and future optimization cases. Using different analysis and data mining methods on the database revealed rules, as well as the relative importance of the design variables of a workstation. The generated rules have been used to identify measures to improve the welding gun workstation design.

Machine learning prediction of calorific value of coal based on the hybrid analysis

ABSTRACT As one of the most important indicators of coal, calorific value (CV) not only determines the value of coal product, but also has a significant impact on the further processing and utilization of coal. Traditional methods of obtaining prediction data suffer from a range of problems such as too many input variables, low prediction accuracy of a single analysis method and lack of sensitivity analysis of input variables. In this paper, a novel hybrid analysis was presented to predict the CV of coal. Pearson correlation coefficient (PCC) was used to remove correlations between variables in order to provide a suitable combination of input variables for ML models. The results showed that based on the optimal combination of input variables (ash, Fe, Mg and Na), RF model provided a better regression, better fit and better robustness on the testing set than the other three models when the performance indicators and the number of input variables were considered. In addition, sensitivity analyses of input variables showed the relative importance of individual variables and the way in which each variable affects the output variables. The present work provided novel insights and ideas for understanding the prediction of the CV of coal.

Machine Learning-Based Analysis and Prediction of Unplanned 30-Day Readmissions After Pituitary Adenoma Resection: A Multi-Institutional Retrospective Study With External Validation.

Unplanned readmission after transsphenoidal resection of pituitary adenoma can occur in up to 10% of patients but is unpredictable. To develop a reliable system for predicting unplanned readmission and create a validated method for stratifying patients by risk. Data sets were retrospectively collected from the National Surgical Quality Improvement Program and 2 tertiary academic medical centers. Eight machine learning classifiers were fit to the National Surgical Quality Improvement Program data, optimized using Bayesian parameter optimization and evaluated on the external data. Permutation analysis identified the relative importance of predictive variables, and a risk stratification system was built using the trained machine learning models. Readmissions were accurately predicted by several classification models with an area under the receiving operator characteristic curve of 0.76 (95% CI 0.68-0.83) on the external data set. Permutation analysis identified the most important variables for predicting readmission as preoperative sodium level, returning to the operating room, and total operation time. High-risk and medium-risk patients, as identified by the proposed risk stratification system, were more likely to be readmitted than low-risk patients, with relative risks of 12.2 (95% CI 5.9-26.5) and 4.2 (95% CI 2.3-8.7), respectively. Overall risk stratification showed high discriminative capability with a C-statistic of 0.73. In this multi-institutional study with outside validation, unplanned readmissions after pituitary adenoma resection were accurately predicted using machine learning techniques. The features identified in this study and the risk stratification system developed could guide clinical and surgical decision making, reduce healthcare costs, and improve the quality of patient care by better identifying high-risk patients for closer perioperative management.