Classification Tree Model Research Articles

A machine learning approach to identify correlates of current e-cigarette use in Canada

Aim: Popularity of electronic cigarettes (i.e. e-cigarettes) is soaring in Canada. Understanding person-level correlates of current e-cigarette use (vaping) is crucial to guide tobacco policy, but prior studies have not fully identified these correlates due to model overfitting caused by multicollinearity. This study addressed this issue by using classification tree, a machine learning algorithm. Methods: This population-based cross-sectional study used the Canadian Tobacco, Alcohol, and Drugs Survey (CTADS) from 2017 that targeted residents aged 15 or older. Forty-six person-level characteristics were first screened in a logistic mixed-effects regression procedure for their strength in predicting vaper type (current vs. former vaper) among people who reported to have ever vaped. A 9:1 ratio was used to randomly split the data into a training set and a validation set. A classification tree model was developed using the cross-validation method on the training set using the selected predictors and assessed on the validation set using sensitivity, specificity and accuracy. Results: Of the 3,059 people with an experience of vaping, the average age was 24.4 years (standard deviation = 11.0), with 41.9% of them being female and 8.5% of them being aboriginal. There were 556 (18.2%) current vapers. The classification tree model performed relatively well and suggested attraction to e-cigarette flavors was the most important correlate of current vaping, followed by young age (< 18) and believing vaping to be less harmful to oneself than cigarette smoking. Conclusions: People who vape due to flavors are associated with very high risk of becoming current vapers. The findings of this study provide evidence that supports the ongoing ban on flavored vaping products in the US and suggests a similar regulatory intervention may be effective in Canada.

Polarized Classification Tree Models: Theory and Computational Aspects

In this paper, a new approach in classification models, called Polarized Classification Tree model, is introduced. From a methodological perspective, a new index of polarization to measure the goodness of splits in the growth of a classification tree is proposed. The new introduced measure tackles weaknesses of the classical ones used in classification trees (Gini and Information Gain), because it does not only measure the impurity but it also reflects the distribution of each covariate in the node, i.e., employing more discriminating covariates to split the data at each node. From a computational prospective, a new algorithm is proposed and implemented employing the new proposed measure in the growth of a tree. In order to show how our proposal works, a simulation exercise has been carried out. The results obtained in the simulation framework suggest that our proposal significantly outperforms impurity measures commonly adopted in classification tree modeling. Moreover, the empirical evidence on real data shows that Polarized Classification Tree models are competitive and sometimes better with respect to classical classification tree models.

Establishing Classification Tree Models in Rheumatoid Arthritis Using Combination of Matrix-Assisted Laser Desorption/Ionization Time-of-Flight Mass Spectrometry and Magnetic Beads.

Background: There is no simple method for early diagnosis and evaluation of rheumatoid arthritis (RA). This study aimed to determine potential biomarkers and establish diagnostic patterns for RA using proteomic fingerprint technology combined with magnetic beads.Methods: The serum protein profiles of 97 RA patients and 76 healthy controls (HCs) were analyzed by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF-MS) with weak cationic exchange (WCX) magnetic beads. Samples were randomly divided into training (83 RA patients and 56 HCs) and test sets (14 RA patients and 20 HCs). Patients were classified according to their Disease Activity Score: in remission, n = 28; with low disease activity, n = 17; with moderate disease activity, n = 21; with high disease activity, n = 31. There are 44 RA patients alone, 22 RA patients with interstitial lung disease (RA-ILD), 18 RA patients with secondary Sjögren's syndrome (RA-sSS), 6 RA patients with osteonecrosis of the femoral head (RA-ONFH), and 7 RA patients with other complications. Eleven patients were treated with etanercept only for half a year, after which their serum protein profiles were detected. The proteomic pattern was identified by Biomarker Patterns Software, and the potential biomarkers for RA diagnosis were further identified and quantified by enzyme-linked immunosorbent assay.Results: The diagnostic pattern with four potential protein biomarkers, mass-to-charge (m/z) 3,448.85, 4,716.71, 8,214.29, and 10,645.10, could accurately recognize RA patients from HCs (specificity, 91.57%; sensitivity, 92.86%). The test set were correctly classified by this model (sensitivity, 95%; specificity, 100%). The components containing the four biomarkers were preliminarily retrieved through the ExPasy database, including the C-C motif chemokine 24 (CCL24), putative metallothionein (MT1DP), sarcolipin (SLN), and C-X-C motif chemokine 11 (CCXL11). Only the CCL24 level was detected to have a significant decrease in the serum of RA patients as compared with HCs (p < 0.05). No significant difference was found in others, but a decreasing trend consistent with the down-regulation of the four biomarkers detected by MALDI-TOF-MS was observed. The diagnostic models could effectively discriminate between RA alone and RA with complications (RA-ILD: m/z 10,645.10 and 12,595.86; RA-sSS: m/z 6,635.62 and 33,897.72; RA-ONFH: m/z 2,071.689). The classification model, including m/z 1,130.776, 1,501.065, 2,091.198, and 11,381.87, could distinguish between RA patients with disease activity and those in remission. RA with low disease activity could be efficiently discriminated from other disease activity patients by specific protein biomarkers (m/z 2,032.31, 2,506.214, and Z9286.495). Two biomarkers (m/z 2,032.31 and 4,716.71) were applied to build the classification model for RA patients with moderate and high disease activities. Biological markers for etanercept (m/z 2,671.604064, 5,801.840579, 8,130.195641, and 9,286.49499) were observed between the responder (n = 7) and non-responder groups (n = 4) (p < 0.05).Conclusion: We successfully established a series of diagnostic models involving RA and RA with complications as well as assessed disease activity. Furthermore, we found that CCL24 may be a valuable auxiliary diagnostic indicator for RA. These results provide reference values for clinical practice in the future.

Early prognostic factors of patients with acquired pneumonia under the analysis of autoregressive integrated moving average model-based pathogenic infectious influenza virus

In view of the high mortality rate and complications of acquired pneumonia (AP), the improved autoregressive integrated moving average (ARIMA) model was applied to predict the early prognostic factors of patients with AP. First, a multi-factor mixed forecast ARIMA-classification and regression trees (CART) classification tree model was established in this study, and the ARIMA-CART model was used for time series fitting analysis to observe the incidence trend of AP. Finally, a predictive model for the risk and prognosis of elderly patients with AP was constructed. The experimental results proved that the serum creatine kinase index and lactate dehydrogenase index of patients with AP had a resistance effect on their prognosis. Therefore, the serum creatine kinase index and lactate dehydrogenase index of patients with AP should be dealt with in actual medical scenarios. The indexes can be reasonably detected and can effectively improve the prognosis of patients. This had a certain reference for the promotion of ARIMA model in the research of early prognostic factors in patients with AP.

Using classification and regression tree modeling to investigate appetite hormones and proinflammatory cytokines as biomarkers to differentiate bipolar I depression from major depressive disorder.

Altered immunity and metabolic profiles have been compared between bipolar depression (BD) and major depressive disorder (MDD). This study aimed at developing a composite predictor of appetite hormones and proinflammatory cytokines to differentiate BD from MDD. This cross-sectional study enrolled patients with BD and those with MDD aged 20 to 59years and displaying depressive episodes. Clinical characteristics (age, sex, body mass index, and depression severity), cytokines (C-reactive protein, interleukin [IL]-2, IL-6, tumor necrosis factor [TNF]-α, P-selectin, and monocyte chemoattractant protein), and appetite hormones (leptin, adiponectin, ghrelin, and insulin) were assessed as potential predictors using a classification and regression tree (CRT) model for differentiating BD from MDD. The predicted probability of a composite predictor of ghrelin and TNF-α was significantly greater (for BD: area under curve=0.877; for MDD: area under curve=0.914) than that of any one marker (all P>.05) to distinguish BD from MDD. The most powerful predictors for diagnosing BD were high ghrelin and TNF-α levels, whereas those for MDD were low ghrelin and TNF-α levels. A composite predictor of ghrelin and TNF-α driven by CRT could assist in the differential diagnosis of BD from MDD with high specificity. Further clinical studies are warranted to validate our results and to explore underlying mechanisms.

Decision Tree Algorithm Identifies Stroke Patients Likely Discharge Home After Rehabilitation Using Functional and Environmental Predictors

Background and purposeThe importance of environmental factors for stroke patients to achieve home discharge was not scientifically proven. There are limited studies on the application of the decision tree algorithm with various functional and environmental variables to identify stroke patients with a high possibility of home discharge. The present study aimed to identify the factors, including functional and environmental factors, affecting home discharge after stroke inpatient rehabilitation using the machine learning method. MethodThis was a cohort study on data from the maintained database of all patients with stroke who were admitted to the convalescence rehabilitation ward of our facility. In total, 1125 stroke patients were investigated. We developed three classification and regression tree (CART) models to identify the possibility of home discharge after inpatient rehabilitation. ResultsAmong three models, CART model incorporating basic information, functional factor, and environmental factor variables achieved the highest accuracy for identification of home discharge. This model identified FIM dressing of the upper body (score of ≤2 or >2) as the first single discriminator for home discharge. Performing house renovation was associated with a high possibility of home discharge even in patients with stroke who had a poor FIM score in the ability to dress the upper body (≤2) at admission into the convalescence rehabilitation ward. Interestingly, many patients who performed house renovation have achieved home discharge regardless of the degree of lower limb paralysis. ConclusionWe identified the influential factors for realizing home discharge using the decision tree algorithm, including environmental factors, in patients with convalescent stroke.

An exploratory study on the use of different composite magnetic and colour fingerprints in aeolian sediment provenance fingerprinting

There is an urgent need for reliable and cost-effective sediment source tracing techniques for apportioning aeolian sediment (sand dune) sources for guiding the selection of best management practices for wind erosion control. Accordingly, the main aim of this study was to quantify the contributions of aeolian sources to sand dune target sediment samples collected in a case study in central Iran using a source fingerprinting procedure based on low-cost fingerprints comprising colour and magnetic tracers. Colour (RGB), magnetic susceptibility (χlf and χhf) and 13 colour and magnetic indices were measured on 54 aeolian sediment source samples and ten aeolian target sediment samples. Three different composite fingerprints for discriminating and apportioning the aeolian sediment sources were selected based on a combination of statistical tests comprising the Kruskal–Wallis H test (KW-H), discriminant function analysis (DFA), principal component & classification analysis (PCCA), and a general classification & regression tree (GC&RT) model. The Modified MixSIR Bayesian un-mixing model was used to apportion aeolian source contributions using the final composite fingerprints. The composite signatures all suggested that the salt flat plain was the dominant (average 63%, and standard deviation, SD, 5.9%) source of the aeolian target sediment samples, whilst agricultural land was second (average 63%, SD 5.6%,) most important. The root mean square difference between the apportionment results based on the three composite fingerprints ranged from 0.2% to 8.3%. Pairwise comparisons of the posterior distributions for the predicted source proportions generated using the three composite signatures showed that eight of 12 pairwise comparisons were not significantly different. Virtual mixture accuracy tests of the fingerprinting models using the three composite signatures suggested errors ranging between 2.2%−20.6% (with a mean of 9.9%), 1.4%−17.0% (mean value 8.3%), and 0.03%−1.0% (mean value 0.8%). The results support the use of low-cost colour and magnetic tracers by investigations into aeolian sediment provenance.

Ensemble-based machine learning approach for improved leak detection in water mains

Abstract This paper presents an acoustic leak detection system for distribution water mains using machine learning methods. The problem is formulated as a binary classifier to identify leak and no-leak cases using acoustic signals. A supervised learning methodology has been employed using several detection features extracted from acoustic signals, such as power spectral density and time-series data. The training and validation data sets have been collected over several months from multiple cities across North America. The proposed solution includes a multi-strategy ensemble learning (MEL) using a gradient boosting tree (GBT) classification model, which has performed better in maximizing detection rate and minimizing false positives as compared with other classification models such as KNN, ANN, and rule-based techniques. Further improvements have been achieved using a multitude of GBT classifiers combined in a parallel ensemble method called bagging algorithm. The proposed MEL approach demonstrates a significant improvement in performance, resulting in a reduction of false positives reports by an order of magnitude.

Individual thermal comfort prediction using classification tree model based on physiological parameters and thermal history in winter

Individual thermal comfort models based on physiological parameters could improve the efficiency of the personal thermal comfort control system. However, the effect of thermal history has not been fully addressed in these models. In this study, climate chamber experiments were conducted in winter using 32 subjects who have different indoor and outdoor thermal histories. Two kinds of thermal conditions were investigated: the temperature dropping (24–16 °C) and severe cold (12 °C) conditions. A simplified method using historical air temperature to quantify the thermal history was proposed and used to predict thermal comfort and thermal demand from physical or physiological parameters. Results show the accuracies of individual thermal sensation prediction was low to about 30% by using the PMV index in cold environments of this study. Base on the sensitivity and reliability of physiological responses, five local skin temperatures (at hand, calf, head, arm and thigh) and the heart rate are optimal input parameters for the individual thermal comfort model. With the proposed historical air temperature as an additional input, the general accuracies using classification tree model C5.0 were increased up by 15.5% for thermal comfort prediction and up by 29.8% for thermal demand prediction. Thus, when predicting thermal demands in winter, the factor of thermal history should be considered.

Spatiotemporal variations of global terrestrial vegetation climate potential productivity under climate change

Evaluating the climate potential productivity (CPP) of terrestrial vegetation is crucial to ascertain the threshold of vegetation productivity, to maximize the utilization of regional climate resources, and to fully display the productivity application level. In this study, the maximum net primary productivity (NPPmax) representing the highest possible productivity of vegetation was calculated using the FLUXNET maximum gross primary productivity (GPPmax) from 177 flux towers. The relationships between NPPmax and a set of climate variables were established using the classification and regression tree (CART) modeling framework. The CART algorithm was used to upscale the CPP to the global scale under the current climate baseline (1980–2018) and future climate scenarios. The spatiotemporal variations in CPP over the globe were analyzed and the impacts of climate factors on it were assessed. The results indicate that global CPPs range from 0 to 2000 g C/m2. The tropical rainforest area is the region with the highest CPP, whereas the lowest CPP occurs in arid/semiarid areas. These two regions were identified as the areas with the largest CPP reductions in the future. The findings reveal that CPP shows signs of productivity saturation and that future climate is not conducive to the increases in vegetation productivity in these regions. The increases in average annual temperature, minimum temperature, and solar radiation are beneficial to CPP increase in most parts of the globe under climate change. However, the negative contribution of maximum temperature increase and precipitation reduction to CPP is higher than the positive contribution of the above three rising factors to CPP in tropical and arid/semiarid areas. Our study is important to aid in creating targeted policies for future sustainable development, resource allocation, and vegetation management.

A classification approach for urban metabolism using the CART model and its application in China

Exploring resource utilization strategies for different types of cities is essential for achieving the sustainable development of cities. Urban metabolism is a systems-oriented approach to understanding the interrelationship between urban socioeconomic status and natural environment through resource utilization. Recognition and identification of urban metabolic types could help in formulating targeted urban resource utilization strategies. However, few studies have investigated the classification of urban metabolism. Therefore, this study combines material flow analysis with the classification and regression tree model to present a classification approach for urban metabolism that considers cities’ socioeconomic, demographic, and climatic conditions and resource consumption. This approach was tested in Chinese prefecture-level cities. The results showed that gross domestic product (GDP) per capita, population, population density, and climate types were significantly correlated with cities’ resource consumption indicators. GDP per capita and climate types were the most important threshold variables for classifying urban resource consumption levels. GDP per capita was the optimal split node for electricity, total energy, water, industrial minerals and metals, and total outputs, while climate types was the optimal split node for biomass, fossil fuels, and construction materials. The level of urban resource consumption in temperate climates was higher than in other climate types. Eight types of urban metabolism were identified in China that were closely related to urban socioeconomic conditions, scale grade, resource endowment, industrial level, and energy structure. The findings of this study provide a reference for determining the problem of urban critical resource consumption and formulating appropriate urban resource utilization strategies.

Association between social capital and loneliness among older adults: a cross-sectional study in Anhui Province, China

BackgroundWe aimed to examine the association between social capital and loneliness in Anhui Province, China.MethodsData were collected from a cross-sectional study using a multi-stage stratified cluster sampling strategy. Data on demographic characteristics, socioeconomic factors, social capital, and loneliness in 1810 older adults (aged 60 years and older) were used for analysis. Binary logistic regression models and a classification and regression tree model were performed to assess the association of social capital and loneliness.ResultsOur results indicated that social capital in terms of lower level of social participation (AOR = 1.38; 95% CI: 1.10–1.74), social connection (AOR = 1.51; 95% CI: 1.18–1.93), and reciprocity (AOR = 1.47; 95% CI: 1.13–1.90) were associated with higher odds of developing loneliness. We noted the interactive effect of different social capital dimensions on loneliness, suggesting that the risk for suffering loneliness was greatest in older people limited in functional ability, with less trust, less social connection, and less social participation.ConclusionsOur findings show that social capital is associated with loneliness in older adults. This implies that social capital, especially in terms of trust, social connection, and social participation may be significant for alleviating loneliness in later life.

Rating prediction of peer-to-peer accommodation through attributes and topics from customer review

This study aims to predict customers’ behavior in classifying their reviews as high rated or low rated using associated attributes and topics found in the review. Knowing customer reviewing action better can lead to a successful strategy implementation of the relevant parties related to this study such as policy to manage customer reviews by keeping their satisfaction high. We applied a big data approach on a dataset of 55,377 reviews from Airbnb listings in the top 10 most visited cities in Indonesia (based on foreign arrivals data). We used The Classification and Regression Tree Model, Random Forest Model, Least Absolute Shrinkage and Selection Operation and Logistic Regression Model, Artificial Neural Network as well as Multi-Layer Perceptron to make prediction’s classification. Those models are used to identify a set of attributes and topics that will increase the chance of the review to render a high rate and a different set of attributes and topics that will lead the review to be low rated. This study found; first, attributes and topics that influence customers' odds to classify their review as high rated or low rated adhere to the understanding of Peer to Peer accommodation attributes. Second, successfully proved that customer reviews' attributes and topics could be used to predict the classification of ratings in Peer to Peer accommodation. Where for Topics, we can predict the rating using Random Forest yields 60.09% accuracy, slightly better than Artificial Neural Network (58.33%) and Multi-Layer Perceptron (58.8%). However, it seems better to use Attributes to predict the rating, where the accuracy is yielded better by applying Artificial Neural Network with 84.79% accuracy compared to Multi-Layer Perceptron with only 72.35% of accuracy.

Analysing the predictive capacity and dose-response of wellness in load monitoring

ABSTRACT This study aimed to identify the predictive capacity of wellness questionnaires on measures of training load using machine learning methods. The distributions of, and dose–response between, wellness and other load measures were also examined, offering insights into response patterns. Data (n= 14,109) were collated from an athlete management systems platform (Catapult Sports, Melbourne, Australia) and were split across three sports (cricket, rugby league and football) with data analysis conducted in R (Version 3.4.3). Wellness (sleep quality, readiness to train, general muscular soreness, fatigue, stress, mood, recovery rating and motivation) as the dependent variable, and sRPE, sRPE-TL and markers of external load (total distance and m.min−1) as independent variables were included for analysis. Classification and regression tree models showed high cross-validated error rates across all sports (i.e., > 0.89) and low model accuracy (i.e., < 5% of variance explained by each model) with similar results demonstrated using random forest models. These results suggest wellness items have limited predictive capacity in relation to internal and external load measures. This result was consistent despite varying statistical approaches (regression, classification and random forest models) and transformation of wellness scores. These findings indicate practitioners should exercise caution when interpreting and applying wellness responses.

Classification and regression trees for predicting the risk of a negative test result for tuberculosis infection in Brazilian healthcare workers: a cross-sectional study.

Healthcare workers (HCWs) have a high risk of acquiring tuberculosis infection (TBI). However, annual testing is resource-consuming. We aimed to develop a predictive model to identify HCWs best targeted for TBI screening. We conducted a secondary analysis of previously published results of 708 HCWs working in primary care services in five Brazilian State capitals who underwent two TBI tests: tuberculin skin test and Quantiferon®-TB Gold in-tube. We used a classification and regression tree (CART) model to predict HCWs with negative results for both tests. The performance of the model was evaluated using the receiver operating characteristics (ROC) curve and the area under the curve (AUC), cross-validated using the same dataset. Among the 708 HCWs, 247 (34.9%) had negative results for both tests. CART identified that physician or a community health agent were twice more likely to be uninfected (probability = 0.60) than registered or aid nurse (probability = 0.28) when working less than 5.5 years in the primary care setting. In cross validation, the predictive accuracy was 68% [95% confidence interval (95%CI): 65 - 71], AUC was 62% (95%CI 58 - 66), specificity was 78% (95%CI 74 - 81), and sensitivity was 44% (95%CI 38 - 50). Despite the low predictive power of this model, CART allowed to identify subgroups with higher probability of having both tests negative. The inclusion of new information related to TBI risk may contribute to the construction of a model with greater predictive power using the same CART technique.

Classification and regression tree (CART) modelling for analysis of shear strength of FRP-RC members

The shear behaviour of concrete members that are reinforced with fibre-reinforced polymer (FRP) bars varies from that of steel reinforced concrete members. The use of non-parametric techniques, which can explain the relationship between different properties of FRP members in more detail, has not yet been well researched in this area. This study utilized a non-parametric technique, namely Classification and Regression Tree (CART) to predict the shear capacity of FRP reinforced concrete members. The members were only reinforced lengthways without vertical reinforcement. A total of 216 experimental results are used to train and test the CART model. The outcomes from CART were compared to traditional models and equations, as well as another non-parametric model from previous research. The comparison showed that CART has better accuracy than other models. It can therefore be used to develop guidelines for the design of FRP concrete members.

Satellite Flood Inundation Assessment and Forecast Using SMAP and Landsat.

The capability and synergistic use of multisource satellite observations for flood monitoring and forecasts is crucial for improving disaster preparedness and mitigation. Here, surface fractional water cover (FW) retrievals derived from Soil Moisture Active Passive (SMAP) L-band (1.4 GHz) brightness temperatures were used for flood assessment over southeast Africa during the Cyclone Idai event. We then focused on five subcatchments of the Pungwe basin and developed a machine learning based approach with the support of Google Earth Engine for daily (24-h) forecasting of FW and 30-m inundation downscaling and mapping. The Classification and Regression Trees model was selected and trained using retrievals derived from SMAP and Landsat coupled with rainfall forecasts from the NOAA Global Forecast System. Independent validation showed that FW predictions over randomly selected dates are highly correlated (R = 0.87) with the Landsat observations. The forecast results captured the flood temporal dynamics from the Idai event; and the associated 30-m downscaling results showed inundation spatial patterns consistent with independent satellite synthetic aperture radar observations. The data-driven approach provides new capacity for flood monitoring and forecasts leveraging synergistic satellite observations and big data analysis, which is particularly valuable for data sparse regions.

Scale effects on the prediction of rare events in mature second-growth oak forests: a simulation study of cavity trees

The abundance of cavity trees varies greatly due to the stochastic nature of cavity formation processes and involved disturbance agents. At small spatial scales such as a stand or plot, cavity tree abundance is extraordinarily difficult to predict precisely using tree and stand factors. In this study we used resampling methods to simulate the effect of spatial scale on cavity tree density (CTD) estimation using cavity tree data collected from a long-term forest experimental project. More than 53,000 measured trees were randomly divided into two approximately equal parts: the construction and test datasets, to construct classification and regression tree (CART) and logistic regression (LR) models to predict cavity probability and to test the accuracy of CTD estimation across varying spatial scales, respectively. Simulation results showed that when the spatial scale was

Development of an Extreme Gradient Boosting Model Integrated With Evolutionary Algorithms for Hourly Water Level Prediction

The establishment of reliable water level prediction models is vital for urban flood control and planning. In this paper, we develop hybrid models (GA-XGBoost and DE-XGBoost) that couple two evolutionary models, a genetic algorithm (GA) and a differential evolution (DE) algorithm, with the extreme gradient boosting (XGBoost) model for hourly water level prediction. The Jungrang urban basin located on the Han River, South Korea, was selected as a case study for the proposed models. Hourly rainfall and water level data were collected between 2003 and 2020 to construct and evaluate the performance of the selected models. To compare the prediction efficiency, two other tree-based models were chosen: classification and registration tree (CART) and random forest (RF) models. A comparison of the results showed that two hybrid models, GA-XGBoost and DE-XGBoost, outperformed RF and CART in the multistep-ahead prediction of water level, and the relative errors of the hybrid model ranged from [2.18%-9.21%], compared to [3.76%-10.41%] and [2.99%-11.88%] for the RF and CART, respectively. Reliable performance was also supported by other measures. In general, the GA-XGBoost and DE-XGBoost models displayed relatively similar performance despite their small differences. The CART model was not preferable for multistep-ahead water level predictions, even though it yielded the lowest Akaike information criterion (AIC) value. This study verifies that despite having some drawbacks when considering long step-ahead prediction and model complexity, hybrid XGBoost models might be superior to many existing models for hourly water level prediction.

Prescriptions of analgesics and anti-inflammatory drugs in municipalities from a Brazilian Southeast state.

The objective of this study was to describe dental prescriptions of non-steroidal anti-inflammatory drugs (NSAID), opioids, and analgesics dispensed by the Brazilian National Health System (BNHS, SUS in Portuguese) of a Southeastern state from January to December 2017, and to analyze their association with socioeconomic and oral health care services' characteristics at municipal level. Data were collected from the Brazilian Integrated Pharmaceutical Care Management System. Medicines were grouped according to the Anatomical Therapeutic Chemical Classification System. The total number of Defined Daily Doses (DDD) and DDD per 1,000 inhabitants (inhab.) per year were presented and compared between groups of municipalities. Data analysis used the Classification and Regression Tree model performed with IBM SPSS 25.0. The total number of NSAID, opioids, and analgesics prescriptions was 70,747 and accounted for 354,221.13 DDD. The most frequently prescribed medicine was ibuprofen (n = 24,676; 34.88%). The number of dental practitioners in the BNHS per 1,000 inhab. (p < 0.001), first dental appointment coverage (p = 0.010), oral health teams per 1,000 inhab. (p=0.022), and the proportion of rural population (p = 0.014) were variables positively associated with the number of DDD of NSAID per 1,000 inhab. per year. Bolsa Família program coverage per 1,000 inhab. (p = 0.022) was negatively associated with NSAID prescription. Regarding analgesics, first dental appointment coverage (p=0.002) and Bolsa Família program coverage per 1,000 inhab. (p = 0.012) were positively associated with DDD per 1,000 inhab. per year. In conclusion, dental prescriptions of analgesics and NSAID in the BNHS were associated with socioeconomic and oral health care services' characteristics.