Quick Unbiased Efficient Statistical Tree Research Articles

Integrating the valence theory and the norm activation theory to understand consumers’ e-waste recycling intention

Electrical and electronic waste (e-waste) is a growing challenge, matching the widespread boom in the use of information and communication technology. Opposite to an alarming increasing amount of e-waste, a low rate of consumer engagement in ensuring the proper disposal of such materials intensifies the pressure on the existing e-waste crisis. To deal with this thorny problem, it is of great interest to grasp consumers’ disposal and recycling behavioral intentions. Therefore, this study attempts to understand complementary perspectives around consumers’ e-waste recycling intention based on the integration of the valence theory and the norm activation theory. Four data mining models using classification and prediction-based algorithms, namely Chi-squared automatic interaction detector (CHAID), Neural network, Discriminant analysis, and Quick, unbiased, efficient statistical tree (QUEST), were employed to analyze a set of the 398 data collected in Vietnam. The results revealed that the social support value is by far the most critical predictor, followed by the utilitarian value, task difficulty, and monetary risk. It is also noteworthy that the awareness of consequences, education background, the ascription of responsibility, and age were also ranked as critical affecting factors. The lowest influential predictors found in this study were income and gender. In addition, a comparison was made in terms of the classification performance of the four utilized data mining techniques. Based on several evaluation measurements (confusion matrix, accuracy, precision, recall, specificity, F-measure, ROC curve, and AUC), the aggregated results suggested that CHAID and Neural network performed the best. The findings of this research are expected to assist policymakers and future researchers in updating all information surrounding consumer behavioral intention-related topics focusing on e-waste. Furthermore, the adoption of data mining algorithms for prediction is another insight of this study, which may shed the light on data mining applications in such environmental studies in the future.

A comparative assessment of gully erosion spatial predictive modeling using statistical and machine learning models

In recent years, gully erosion has ceased many development activities and imposed a living threat to local communities residing in southern Iran. Hence, this study sets out to investigate the prediction performance of a machine learning model named the quick, unbiased, efficient statistical tree (QUEST) model for gully susceptibility mapping. Its results were compared to two conventional statistical models: frequency ratio (FR) and evidential belief function (EBF). The area under the receiver operating characteristic (AUROC) and the true skill statistic (TSS) metrics were adopted to assess models' goodness-of-fit and predictive performance in the corresponding training and validation stages. Results revealed that the QUEST model outperforms its counterparts by giving respective AUROC and TSS values of 88.5% and 0.77 in the training stage, followed by EBF (82.3% and 0.65) and FR (80.4% and 0.62). Similarly, the QUEST model showed the highest AUROC and TSS values in the validation stage (83.2% and 0.63, respectively), followed by the EBF (78.6% and 0.63, respectively) and FR (77.1% and 0.58, respectively). Further scrutinization attested that the QUEST model offers a more practical, compendious, and adaptable susceptibility map based on which about 32% of the study area was identified as the high susceptibility zone to gully erosion. Hence, highly gully susceptible areas require pragmatic mitigation plans. In addition, the application of machine learning models for gully erosion merits further studies.

Predicting Entrepreneurial Intentions among the Youth in Serbia with a Classification Decision Tree Model with the QUEST Algorithm

Youth unemployment rates present an issue both in developing and developed countries. The importance of analyzing entrepreneurial activities comes from their significant role in economic development and economic growth. In this study, a 10-year research was conducted. The dataset included 5670 participants—students from Serbia. The main goal of the study is to attempt to predict entrepreneurial intentions among the Serbian youth by analyzing demographics characteristics, close social environment, attitudes, awareness of incentive means, and environment assessment as potential influencing factors. The data analysis included Chi-square, Welch’s t-test, z-test, linear regression, binary logistic regression, ARIMA (Autoregressive Integrated Moving Average) regression, and a QUEST (Quick, Unbiased, Efficient, Statistical Tree) classification tree algorithm. The results are interesting and indicate that entrepreneurial intentions can be partially predicted using the dataset in this current study. Further, most likely due to the robust dataset, the results are not complementary with similar studies in this domain; therefore, these findings expand the current literature and invite future research.

Measuring financial performance of Indian manufacturing firms: application of decision tree algorithms

PurposeDetermining the relevant information using financial measures is of great interest for various stakeholders to analyze the performance of the firm. This paper aims at identifying these financial measures (ratios) which critically affect the firm performance. The authors specifically focus on discovering the most prominent ratios using a two-step process. First, the authors use an exploratory factor analysis to identify the underlying dimensions of these ratios, followed by predictive modeling techniques to identify the potential relationship between measures and performance.Design/methodology/approachThe study uses data of 25 financial variables for a sample of 1923 Indian manufacturing firms which exist continuously between 2011 and 2018. For prediction models, four popular decision tree algorithms [Chi-squared automatic interaction detector (CHAID), classification and regression trees (C&RT), C5.0 and quick, unbiased, efficient statistical tree (QUEST)] were investigated, and the information fusion-based sensitivity analyses were performed to identify the relative importance of these input measures.FindingsResults show that C5.0 and CHAID algorithms produced the best predictive results. The fusion sensitivity results find that net profit margin and total assets turnover rate are the most critical factors determining the firm performance in an Indian manufacturing context. These findings may enable managers in their decision-making process and also have vital implications for investors in assessing the performance of the firm.Originality/valueTo the best of the authors’ knowledge, the current paper is the first to address the application of decision tree algorithms to predict the performance of manufacturing firms in an emerging economy such as India, with the latest data. This practical perspective helps the organizations in managing the critical parameters for the firm’s growth.

Classification of transformed anchovy products based on the use of element patterns and decision trees to assess traceability and country of origin labelling

Quadrupole inductively coupled plasma mass spectrometry (Q-ICP-MS) and direct mercury analysis were used to determine the elemental composition of 180 transformed (salt-ripened) anchovies from three different fishing areas before and after packaging. To this purpose, four decision trees-based algorithms, corresponding to C5.0, classification and regression trees (CART), chi-squareautomatic interaction detection (CHAID), and quick unbiased efficient statistical tree (QUEST) were applied to the elemental datasets to find the most accurate data mining procedure to achieve the ultimate goal of fish origin prediction. Classification rules generated by the trained CHAID model optimally identified unlabelled testing bulk anchovies (93.9% F-score) by using just 6 out of 52 elements (As, K, P, Cd, Li, and Sr). The finished packaged product was better modelled by the QUEST algorithm which recognised the origin of anchovies with F-score of 97.7%, considering the information carried out by 5 elements (B, As, K. Cd, and Pd). Results obtained suggested that the traceability system in the fishery sector may be supported by simplified machine learning techniques applied to a limited but effective number of inorganic predictors of origin.

Go/No-Go Decision Model for Owners Using Exhaustive CHAID and QUEST Decision Tree Algorithms

Go/no-go execution decisions are one of the most important strategic decisions for owners during the early stages of construction projects. Restructuring the process of decision-making during these early stages may have sustainable results in the long run. The purpose of this paper is to establish proper go/no-go decision-tree models for owners. The decision-tree models were developed using Exhaustive Chi-square Automatic Interaction Detector (Exhaustive CHAID) and Quick, Unbiased, Efficient Statistical Tree (QUEST) algorithms. Twenty-three go/no-go key factors were collected through an extensive literature review. These factors were divided into four main risk categories: organizational, project/technical, legal, and financial/economic. In a questionnaire distributed among the construction professionals, the go/no-go variables were asked to be ranked according to their perceived significance. Split-sample validation was applied for testing and measuring the accuracy of the Exhaustive CHAID and QUEST models. Moreover, Spearman’s rank correlation and analysis of variance (ANOVA) tests were employed to identify the statistical features of the 100 responses received. The result of this study benchmarks the current assessment models and develops a simple and user-friendly decision model for owners. The model is expected to evaluate anticipated risk factors in the project and reduce the level of uncertainty. The Exhaustive CHAID and QUEST models are validated by a case study. This paper contributes to the current body of knowledge by identifying the factors that have the biggest effect on an owner’s decision and introducing Exhaustive CHAID and QUEST decision-tree models for go/no-go decisions for the first time, to the best of the authors’ knowledge. From the “sustainability” viewpoint, this study is significant since the decisions of the owner, based on a rigorous model, will yield sustainable and efficient projects.

Comparing the Balanced Accuracy of Deep Neural Network and Machine Learning for Predicting the Depressive Disorder of Multicultural Youth

Multicultural youth are more likely to experience negative emotions (e.g. depressive symptoms) due to social prejudice and discrimination. Nevertheless, previous studies that analyzed the emotional aspects of multicultural youth mainly compared the characteristics of multicultural youth and those of the other youth or identified individual risk factors using a regression model. This study developed models to predict the depressive disorders of multicultural youth based on the Quick Unbiased Efficient Statistical Tree (QUEST), Classification And Regression Trees (CART), gradient boosting machine (G-B-M), random forest, and deep neural network (deep-NN) using epidemiological data representing multicultural youth and compared the prediction performance (PRED PER) of the developed models. Our study analyzed 19,431 youths (9,835 males and 9,596 females) aged between 19 and 24 years old. We developed models for predicting the self-awareness of health of youths by using QUEST, CART, G-B-M, random forest, and deep-NN and compared the balanced accuracy of them to evaluate their PRED PER. Among 19,431 subjects, 42.9% (5,838 people) experienced a depressive disorder in the past year. The results of our study confirmed that deep-NN had the best PRED PER with a specificity of 0.85, a sensitivity of 0.71, and a balanced accuracy of 0.78. It will be necessary to develop a model with optimal PRED PER by tuning hyperparameters (e.g., number of hidden layers, number of iterations, and activation function, number of hidden nodes) of deep-NN.

Flood Risk Mapping of Jakarta Using Genetic Algorithm Rule-Set Production (GARP) and Quick Unbiased Efficient Statistical Tree (QUEST) Methods

Jakarta as the country’s capital, has experienced a series of floods that crippled cities in 2002, 2007 and 2013 and extreme rainfall has always been the main cause of major flood with casualties and property. Given the increasing impact of flooding in Jakarta, methods for assessing current and future flood risks is needed. A model is needed that can predict and determine which areas have the potential to be affected by future floods. Several models have been carried out for prediction and assessment of flood risk in Jakarta and several other major cities, namely GIS based on SVM models, hierarchical process analysis models (AHP), machine learning with ensemble models, and multivariate discriminant, classification, and regression trees. However, from several methods there are still some weaknesses, one of which is the time to get a predictive model that is quite long, which is 3-4 hours depending on the amount of data. To overcome this problem, genetic algorithm rule-set production (GARP) and quick unbiased efficient statistical tree (QUEST) models can create a prediction model with a shorter time with the same results. Several factors that influence flooding are used as input model data: precipitation, slope, distance to river, distance to channel, depth to groundwater, land use, elevation, and flood data for 2002 - 2019. The Area under receiver-operator characteristic curve (AUC-ROC) and root mean square error (RMSE) were used as evaluations for model performance. The results showed that the GARP model (AUC-ROC = 94%, RMSE = 0.2) had higher performance accuracy than the QUEST model (AUC-ROC = 84%, RMSE = 0.4). The results also indicated that the distance to channel, land use, and elevation as important parameters in determining flood hazards.

A Modeling Comparison of Groundwater Potential Mapping in a Mountain Bedrock Aquifer: QUEST, GARP, and RF Models

In some arid regions, groundwater is the only source of water for human needs, so understanding groundwater potential is essential to ensure its sustainable use. In this study, three machine learning models (Genetic Algorithm for Rule-Set Production (GARP), Quick Unbiased Efficient Statistical Tree (QUEST), and Random Forest (RF)) were applied and verified for spatial prediction of groundwater in a mountain bedrock aquifer in Piranshahr Watershed, Iran. A spring location dataset consisting of 141 springs was prepared by field surveys, and from this three different sample datasets (S1–S3) were randomly generated (70% for training and 30% for validation). A total of 10 groundwater conditioning factors were prepared for modeling, namely slope percent, relative slope position (RSP), plan curvature, altitude, drainage density, slope aspect, topographic wetness index (TWI), terrain ruggedness index (TRI), land use, and lithology. The area under the receiver operating characteristic curve (AUC) and true skill statistic (TSS) were used to evaluate the accuracy of models. The results indicated that all models had excellent goodness-of-fit and predictive performance, but that RF (AUCmean = 0.995, TSSmean = 0.89) and GARP (AUCmean = 0.957, TSSmean = 0.82) outperformed QUEST (AUCmean = 0.949, TSSmean = 0.74). In robustness analysis, RF was slightly more sensitive than GARP and QUEST, making it necessary to consider several random partitioning options for preparing training and validation groups. The outcomes of this study can be useful in sustainable management of groundwater resources in the study region.

The predictive power of game-related statistics for the final result under the rule changes introduced in the men’s world water polo championship: a classification-tree approach

ABSTRACTThe objectives of this study were (i) to compare water polo game-related statistics by match outcome (winning and losing teams) after the application of the new rules, and (ii) to develop a classification tree model explaining the performance in elite men’s water polo. Forty-eight matches that were played in the 18th FINA World Championships were analysed. The dependent variable was match outcome and the independent variables were the game-related statistics. To determine the differences between the winning and losing teams, a parametric (paired-sample t-test) or non-parametric (Wilcoxon signed-rank test) test was applied, depending on whether or not the variable satisfied normality. The effect sizes (ES) of the differences were calculated. In order to determine which variables best predict the final outcome, a decision tree was constructed. This was a tree based on the supervised learning method called QUEST (Quick, Unbiased, Efficient, Statistical Tree). Ten variables differentiated between winning and losing teams (ES ≥ 0.80): four were related to the effectiveness of throwing, three to the effectiveness of the goalkeeper, and three to other actions. The decision tree correctly classified 83.9% of the teams with the variables GB shots, actions goals, time-outs, and steals.

Evaluation of CART, CHAID, and QUEST algorithms: a case study of construction defects in Taiwan

ABSTRACTThe type and number of defects constitute a major indicator of project quality and are thus emphasized in project management. Therefore, it is necessary to explore appropriate tools and methods to train/test/analyze defect-related big data in order to effectively explain the cause/rule/importance of the defect, to understand the focus of site management, and to effectively prevent defects. The aim of this study is to explore the capability of three kinds of decision tree algorithms, namely classification and regression tree (CART), chi-squared automatic interaction detection (CHAID) and quick unbiased efficient statistical tree algorithms (QUEST), in predicting the construction project grade given defects. Firstly, a total of 499 types of defects were identified after the analysis of the data of 990 projects obtained from the Public Construction Management Information System (PCMIS). Secondly, inspection scores and defect frequencies were estimated to perform cluster analysis for re-grouping the data to create project grade. Thirdly, decision trees were used to classify rules for defects and project grades. The results revealed that, among the three algorithms, CHAID generated the most classification rules and exhibited the highest defect prediction accuracy. The finding of this research can improve the defect prediction accuracy and management effectiveness for construction industry.

Discriminating Urban Forest Types from Sentinel-2A Image Data through Linear Spectral Mixture Analysis: A Case Study of Xuzhou, East China

Urban forests are an important component of the urban ecosystem. Urban forest types are a key piece of information required for monitoring the condition of an urban ecosystem. In this study, we propose an urban forest type discrimination method based on linear spectral mixture analysis (LSMA) and a support vector machine (SVM) in the case study of Xuzhou, east China. From 10-m Sentinel-2A imagery data, three different vegetation endmembers, namely broadleaved forest, coniferous forest, and low vegetation, and their abundances were extracted through LSMA. Using a combination of image spectra, topography, texture, and vegetation abundances, four SVM classification models were performed and compared to investigate the impact of these features on classification accuracy. With a particular interest in the role that vegetation abundances play in classification, we also compared SVM and other classifiers, i.e., random forest (RF), artificial neural network (ANN), and quick unbiased efficient statistical tree (QUEST). Results indicate that (1) the LSMA method can derive accurate vegetation abundances from Sentinel-2A image data, and the root-mean-square error (RMSE) was 0.019; (2) the classification accuracies of the four SVM models were improved after adding topographic features, textural features, and vegetation abundances one after the other; (3) the SVM produced higher classification accuracies than the other three classifiers when identical classification features were used; and (4) vegetation endmember abundances improved classification accuracy regardless of which classifier was used. It is concluded that Sentinel-2A image data has a strong capability to discriminate urban forest types in spectrally heterogeneous urban areas, and that vegetation abundances derived from LSMA can enhance such discrimination.

Application of classification methods to analyze chemicals in drinking water quality

To analyze drinking water dataset, various statistical methods have been applied, including discriminant analysis, logistic regression and cluster analysis, to construct models for the identification of important input variables. Among them decision trees are more flexible than other statistical classification methods because it provides us a complete path or frame to reach a specific decision with simplicity and ease of understanding about critical variables. This article describes the application of classification decision trees for the analysis of drinking water quality affecting variables and includes discussion about these based on various methods as well as their comparison to reach the best approach for the further analysis about understudy area. In this study, samples of filtered water are taken from 100 pumps located in different union councils of the Lahore city. The classification trees are constructed on the basis of input quality variables, and the results are reported in the form of confusion matrix. Four techniques, including Chi-square Automatic Interaction Detector, Exhaustive Chi-square Automatic Interaction Detector, Classification and Regression Tree and Quick Unbiased Efficient Statistical Tree, were used. Three experiments were conducted to get performance evaluation of the models by the number of misclassified units. The first method used complete dataset, the second one is based on the cross-validation, while the last one is based on the random subsampling.

Urban flood risk mapping using the GARP and QUEST models: A comparative study of machine learning techniques

Flood risk mapping and modeling is important to prevent urban flood damage. In this study, a flood risk map was produced with limited hydrological and hydraulic data using two state-of-the-art machine learning models: Genetic Algorithm Rule-Set Production (GARP) and Quick Unbiased Efficient Statistical Tree (QUEST). The flood conditioning factors used in modeling were: precipitation, slope, curve number, distance to river, distance to channel, depth to groundwater, land use, and elevation. Based on available reports and field surveys for Sari city (Iran), 113 points were identified as flooded areas (with each flooded zone assigned a value of 1). Different conditioning factors, including urban density, quality of buildings, age of buildings, population density, and socio-economic conditions, were taken into account to analyze flood vulnerability. In addition, the weight of these conditioning factors was determined based on expert knowledge and Fuzzy Analytical Network Process (FANP). An urban flood risk map was then produced using flood hazard and flood vulnerability maps. The area under the receiver-operator characteristic curve (AUC-ROC) and Kappa statistic were applied to evaluate model performance. The results demonstrated that the GARP model (AUC-ROC = 93.5%, Kappa = 0.86) had higher performance accuracy than the QUEST model (AUC-ROC = 89.2%, Kappa = 0.79). The results also indicated that distance to channel, land use, and elevation played major roles in flood hazard determination, whereas population density, quality of buildings, and urban density were the most important factors in terms of vulnerability. These findings demonstrate that machine learning models can help in flood risk mapping, especially in areas where detailed hydraulic and hydrological data are not available.

Survival Prediction of Patients with Breast Cancer: Comparisons of Decision Tree and Logistic Regression Analysis

Background: Breast cancer is the first cause of cancer-related deaths among women in Iran. Objectives: The aim of the present study was to compare the traditional statistical analysis and data mining technique as the research methods for identifying the prognostic factors regarding the survival time of patients with breast cancer. Decision tree method is one of the predictive models that used in the medical field. The most used algorithms are classification and regression trees (CART), the quick, unbiased, efficient statistical tree (QUEST), Chi-square automatic interaction detector (CHAIDs) algorithm, and the C5.0 algorithm. Methods: We used data for 438 patients, who were referred to cancer research center in Shahid Beheshti University of Medical Sciences. The patients were visited and treated during 1992 to 2012 and followed up until October 2014. The data were analyzed by regression logistic and decision tree method. Six measures for evaluation of predictive performance of different models were used. Results: The C5.0 algorithm performed better than CHAID, QUEST, CART algorithms, and the logistic regression in predicting breast cancer survival. The multiple logistic regression results indicated that the factors of age at diagnosis, histologic grade, axillary lymph node status, and type of surgery were statistically significant with regard to the probability of death in patients with breast cancer. Moreover, based on C4.5 they reported that tumor size, age of menarche, hormonal therapy, axillary nodal status, and histological grade are the most prominent variables. Conclusions: The more precise methods can identify the more accurate predictors. The decision tree method was able to predict the probability of death more accurately compared with the conventional logistic regression. Some improvements for classical classification tree such as boosting and bagging have been developed in order to obtain better predictive performance. We suggest that the modern classification tree method in the breast cancer context be the focus of future studies.

An Effective Financial Statements Fraud Detection Model for the Sustainable Development of Financial Markets: Evidence from Taiwan

This study aims to establish a rigorous and effective model to detect enterprises’ financial statements fraud for the sustainable development of enterprises and financial markets. The research period is 2004–2014 and the sample is companies listed on either the Taiwan Stock Exchange or the Taipei Exchange, with a total of 160 companies (including 40 companies reporting financial statements fraud). This study adopts multiple data mining techniques. In the first stage, an artificial neural network (ANN) and a support vector machine (SVM) are deployed to screen out important variables. In the second stage, four types of decision trees (classification and regression tree (CART), chi-square automatic interaction detector (CHAID), C5.0, and quick unbiased efficient statistical tree (QUEST)) are constructed for classification. Both financial and non-financial variables are selected, in order to build a highly accurate model to detect fraudulent financial reporting. The empirical findings show that the variables screened with ANN and processed by CART (the ANN + CART model) yields the best classification results, with an accuracy of 90.83% in the detection of financial statements fraud.

Decision‐tree‐based methods for differential diagnosis of β‐thalassemia trait from iron deficiency anemia

AbstractSeveral discrimination indices have been proposed to distinct between β‐thalassemia trait (βTT) and iron deficiency anemia (IDA). This study is the first application of tree‐based methods for differential diagnosis of βTT from IDA. One hundred forty‐four patients with hypochromic microcytic anemia aged 18–40 years old from Ayat Hospital of Tehran were recruited. Classification and Regression tree, CHi‐squared Automatic Interaction Detector (CHAID), Exhaustive CHi‐squared Automatic Interaction Detector (E‐CHAID), Quick, Unbiased, Efficient Statistical Tree (QUEST), Classification Rule with Unbiased Interaction Selection and Estimation (CRUISE), and Generalized, Unbiased, Interaction Detection and Estimation (GUIDE) have been used to discriminate the diagnosis. Mean corpuscular volume (MCV) was found as the main predictor in discrimination. All the mentioned tree‐based methods showed acceptable sensitivity, specificity, accuracy, Youden's index, false positive and negative rate, positive and negative predictive values and AUC in differential diagnosis of βTT from IDA. However, Classification Rule with Unbiased Interaction Selection and Estimation revealed more precise classification with an area under the curve value of 0.99. Decision‐tree‐based methods can be used to develop sensitive and accurate diagnostic methods for differentiating βTT from IDA.

Decision tree-based modelling for identification of potential interactions between type 2 diabetes risk factors: a decade follow-up in a Middle East prospective cohort study

ObjectiveThe current study was undertaken for use of the decision tree (DT) method for development of different prediction models for incidence of type 2 diabetes (T2D) and for exploring interactions...

Classification-based data mining for identification of risk patterns associated with hypertension in Middle Eastern population: A 12-year longitudinal study.

Hypertension is a critical public health concern worldwide. Identification of risk factors using traditional multivariable models has been a field of active research. The present study was undertaken to identify risk patterns associated with hypertension incidence using data mining methods in a cohort of Iranian adult population.Data on 6205 participants (44% men) age > 20 years, free from hypertension at baseline with no history of cardiovascular disease, were used to develop a series of prediction models by 3 types of decision tree (DT) algorithms. The performances of all classifiers were evaluated on the testing data set.The Quick Unbiased Efficient Statistical Tree algorithm among men and women and Classification and Regression Tree among the total population had the best performance. The C-statistic and sensitivity for the prediction models were (0.70 and 71%) in men, (0.79 and 71%) in women, and (0.78 and 72%) in total population, respectively. In DT models, systolic blood pressure (SBP), diastolic blood pressure, age, and waist circumference significantly contributed to the risk of incident hypertension in both genders and total population, wrist circumference and 2-h postchallenge plasma glucose among women and fasting plasma glucose among men. In men, the highest hypertension risk was seen in those with SBP > 115 mm Hg and age > 30 years. In women those with SBP > 114 mm Hg and age > 33 years had the highest risk for hypertension. For the total population, higher risk was observed in those with SBP > 114 mm Hg and age > 38 years.Our study emphasizes the utility of DTs for prediction of hypertension and exploring interaction between predictors. DT models used the easily available variables to identify homogeneous subgroups with different risk pattern for the hypertension.

Association between Prenatal Environmental Factors and Child Autism: A Case Control Study in Tianjin, China

ObjectiveTo investigate the association between autism and prenatal environmental risk factors. MethodsA case-control study was conducted among 193 children with autism from the special educational schools and 733 typical development controls matched by age and gender by using questionnaire in Tianjin from 2007 to 2012. Statistical analysis included quick unbiased efficient statistical tree (QUEST) and logistic regression in SPSS 20.0. ResultsThere were four predictors by QUEST and the logistic regression analysis, maternal air conditioner use during pregnancy (OR=0.316, 95% CI: 0.215-0.463) was the single first-level node (χ2=50.994, P=0.000); newborn complications (OR=4.277, 95% CI: 2.314-7.908) and paternal consumption of freshwater fish (OR=0.383, 95% CI: 0.256-0.573) were second-layer predictors (χ2=45.248, P=0.000; χ2=24.212, P=0.000); and maternal depression (OR=4.822, 95% CI: 3.047-7.631) was the single third-level predictor (χ2=23.835, P=0.000). The prediction accuracy of the tree was 89.2%. ConclusionThe air conditioner use during pregnancy and paternal freshwater fish diet might be beneficial for the prevention of autism, while newborn complications and maternal depression might be the risk factors.