Exhaustive Chi-squared Automatic Interaction Detector Research Articles

Landslide Susceptibility Mapping and Comparison Using Decision Tree Models: A Case Study of Jumunjin Area, Korea

We assessed landslide susceptibility using Chi-square Automatic Interaction Detection (CHAID), exhaustive CHAID, and Quick, Unbiased, and Efficient Statistical Tree (QUEST) decision tree models in Jumunjin-eup, Gangneung-si, Korea. A total of 548 landslides were identified based on interpretation of aerial photographs. Half of the 548 landslides were selected for modeling, and the remaining half were used for verification. We used 20 landslide control factors that were classified into five categories, namely topographic elements, hydrological elements, soil maps, forest maps, and geological maps, to determine landslide susceptibility. The relationships of landslide occurrence with landslide-inducing factors were analyzed using CHAID, exhaustive CHAID, and QUEST models. The three models were then verified using the area under the curve (AUC) method. The results showed that the CHAID model (AUC = 87.1%) was more accurate than the exhaustive CHAID (AUC = 86.9%) and QUEST models (AUC = 82.8%). The verification results showed that the CHAID model had the highest accuracy. There was high susceptibility to landslides in mountainous areas and low susceptibility in coastal areas. Analyzing the characteristics of the landslide control factors in advance will enable us to obtain more accurate results.

Development of Building Thermal Load and Discomfort Degree Hour Prediction Models Using Data Mining Approaches

Thermal load and indoor comfort level are two important building performance indicators, rapid predictions of which can help significantly reduce the computation time during design optimization. In this paper, a three-step approach is used to develop and evaluate prediction models. Firstly, the Latin Hypercube Sampling Method (LHSM) is used to generate a representative 19-dimensional design database and DesignBuilder is then used to obtain the thermal load and discomfort degree hours through simulation. Secondly, samples from the database are used to develop and validate seven prediction models, using data mining approaches including multilinear regression (MLR), chi-square automatic interaction detector (CHAID), exhaustive CHAID (ECHAID), back-propagation neural network (BPNN), radial basis function network (RBFN), classification and regression trees (CART), and support vector machines (SVM). It is found that the MLR and BPNN models outperform the others in the prediction of thermal load with average absolute error of less than 1.19%, and the BPNN model is the best at predicting discomfort degree hour with 0.62% average absolute error. Finally, two hybrid models—MLR (MLR + BPNN) and MLR-BPNN—are developed. The MLR-BPNN models are found to be the best prediction models, with average absolute error of 0.82% in thermal load and 0.59% in discomfort degree hour.

Florida Populations Most at Risk of Not Being Up to Date With Colorectal Cancer Screening.

IntroductionThe purpose of this study was to examine the characteristics of populations at risk of not being up to date on colorectal cancer screening in Florida.MethodsWe used Exhaustive Chi-squared Automatic Interaction Detection, a classification tree analysis, to identify subgroups not up to date with colorectal cancer screening using the 2013 Florida Behavioral Risk Factor Surveillance System. The data set was restricted to adults aged 50 to 75 years (n = 14,756).ResultsOnly 65.5% of the sample was up to date on colorectal cancer screening. Having no insurance and having a primary care provider were the most significant predictors of not being up to date on screening. The highest risk subgroups were 1) respondents with no insurance and no primary care provider, regardless of their employment status (screening rate, 12.1%–23.7%); 2) respondents with no insurance but had a primary care provider and were employed (screening rate, 32.3%); and 3) respondents with insurance, who were younger than 55 years, and who were current smokers (screening rate, 42.0%).ConclusionSome populations in Florida are at high risk for not being up to date on colorectal cancer screening. To achieve Healthy People 2020 goals, interventions may need to be further tailored to target these subgroups.

A Simple Ultrasound Based Classification Algorithm Allows Differentiation of Benign from Malignant Breast Lesions by Using Only Quantitative Parameters

PurposeWe hypothesized that different quantitative ultrasound (US) parameters may be used as complementary diagnostic criteria and aimed to develop a simple classification algorithm to distinguish benign from malignant breast lesions and aid in the decision to perform biopsy or not.ProceduresOne hundred twenty-four patients, each with one biopsy-proven, sonographically evident breast lesion, were included in this prospective, IRB-approved study. Each lesion was examined with B-mode US, Color/Power Doppler US and elastography (Acoustic Radiation Force Impulse–ARFI). Different quantitative parameters were recorded for each technique, including pulsatility (PI) and resistive Index (RI) for Doppler US and lesion maximum, intermediate, and minimum shear wave velocity (SWVmax, SWVinterm, and SWVmin) as well as lesion-to-fat SWV ratio for ARFI. Receiver operating characteristic curve (ROC) analysis was used to evaluate the diagnostic performance of each quantitative parameter. Classification analysis was performed using the exhaustive chi-squared automatic interaction detection method. Results include the probability for malignancy for every descriptor combination in the classification algorithm.ResultsSixty-five lesions were malignant and 59 benign. Out of all quantitative indices, maximum SWV (SWVmax), and RI were included in the classification algorithm, which showed a depth of three ramifications (SWVmax ≤ or > 3.16; if SWVmax ≤ 3.16 then RI ≤ 0.66, 0.66–0.77 or > 0.77; if RI ≤ 0.66 then SWVmax ≤ or > 2.71). The classification algorithm leads to an AUC of 0.887 (95 % CI 0.818–0.937, p < 0.0001), a sensitivity of 98.46 % (95 % CI 91.7–100 %), and a specificity of 61.02 % (95 % CI 47.4–73.5 %). By applying the proposed algorithm, a false-positive biopsy could have been avoided in 61 % of the cases.ConclusionsA simple classification algorithm incorporating two quantitative US parameters (SWVmax and RI) shows a high diagnostic performance, being able to accurately differentiate benign from malignant breast lesions and lower the number of unnecessary breast biopsies in up to 60 % of all cases, avoiding any subjective interpretation bias.

Diagnosis related group grouping study of senile cataract patients based on E-CHAID algorithm.

To figure out the contributed factors of the hospitalization expenses of senile cataract patients (HECP) and build up an area-specified senile cataract diagnosis related group (DRG) of Shanghai thereby formulating the reference range of HECP and providing scientific basis for the fair use and supervision of the health care insurance fund. The data was collected from the first page of the medical records of 22 097 hospitalized patients from tertiary hospitals in Shanghai from 2010 to 2012 whose major diagnosis were senile cataract. Firstly, we analyzed the influence factors of HECP using univariate and multivariate analysis. DRG grouping was conducted according to the exhaustive Chi-squared automatic interaction detector (E-CHAID) model, using HECP as target variable. Finally we evaluated the grouping results using non-parametric test such as Kruskal-Wallis H test, RIV, CV, etc. The 6 DRGs were established as well as criterion of HECP, using age, sex, type of surgery and whether complications/comorbidities occurred as the key variables of classification node of senile cataract cases. The grouping of senile cataract cases based on E-CHAID algorithm is reasonable. And the criterion of HECP based on DRG can provide a feasible way of management in the fair use and supervision of medical insurance fund.

Predicting minor nutrient requirements of hazelnut shoot cultures using regression trees

Tissue culture data is non-linear in nature. Decision tree algorithms stand out in revealing the non-linear interactions and relationships between the predictors and responses. Classification and regression tree (CART), chi squared automatic interaction detector (CHAID) and exhaustive CHAID are the common decision tree algorithms. These three models were employed to predict and optimize the effect of minor mineral nutrients on shoot cultures of Corylus avellana L. cultivars. H3BO3, CuSO4·5H2O, MnSO4·H2O, Na2MoO4·2H2O and Zn(NO3)2·6H2O were tested in a range of 0.5 × to 4 × Driver and Kuniyuki (DKW) medium within a RSM optimal design. NiSO4·6H2O was also an input within the design with varying levels of 0 to 6 µM. Shoot quality and length were affected by genotype, B and Mo amounts. Multiplication rate depended on genotype, B, Zn and Cu levels. Callus formation was affected by genotype and B. Leaf size depended on genotype, Zn and Mn concentrations. Cu was a significant predictor of leaf color and Ni slightly improved SPAD readings (chlorophyll content). CART in general outperformed CHAID and exhaustive CHAID in terms of the predictive performance. Both CHAID and exhaustive CHAID failed to generate a tree model for a leaf size response. The optimal minor nutrients for hazelnuts based on the predictions of the CART algorithm were suggested to be: B 2.3 × DKW, Cu 0.5×, Mn 0.5×, 2 × Mo and Zn 2×.

Team performance indicators that predict match outcome and points difference in professional rugby league

Performance indicators allow for the objective quantification of performance, however, limited PI research for professional rugby league exists. Therefore, this paper assessed 24 relative PIs (home value minus away) from all 27 rounds of the 2012, 2013 and 2014 European Super League seasons, collected by Opta, amounting to 567 matches. Backwards logistic (match outcome) and linear (points difference) regression models were used alongside exhaustive Chi-Square Automatic Interaction Detection decision trees to identify performance indicators (PIs) and key performance indicators. Teams had a higher chance of winning and would gain more points when they scored first (OR = 1.6, β = 2.4) and increased completed sets (OR = 1.2, β = 1.2) by one unit. Conversely, teams had a lower chance of winning when they increased scoots (OR = 0.9, β = −0.2). However, some PIs which were thought to be important (as identified by previous literature) were removed from the analysis thus calling into question the appropriateness of stepwise methods. Future research may consider utilising dimension reduction techniques when analysing large data-sets that encompass multiple variables.

Study on diagnosis-related groups of inpatients′ expenses for respiratory system diseases

Objective To explore diagnosis-related groups(DRGs) case mixes and development approaches for medicare expense standard fitting patients with respiratory system diseases in Sichuan province. Methods 280 717 cases of respiratory system diseases were sampled from the homepages of medical records of general hospitals in Sichuan. These cases were grouped by means of the exhaustive chi-square automatic interaction detector in the decision tree model and the medicare costs standard was derived using the relative-ratio weighting coefficient. Results The main classification nodes of respiratory diseases were age and patient clinical complexity level (PCCL). Patients were classified into 158 disease diagnosis related groups, including 122 DRGs of internal medicine and 36 DRGs in surgical medicine. The max relative-ratio weighting coefficient was 14.04 and the min one was 0.29. And the extreme inpatients′ expenses can affect the identification of classification nodes, calculation of relative weighting coefficient and medicare cost standard. Conclusions Large sample size is advantageous in establishing DRGs and calculating the medicare costs standard based on relative-ratio weighting coefficient. It is however imperative to strengthen monitoring on extreme inpatients′ costs and control the homepage quality of medical records. Key words: Respiratory system diseases; Diagnosis-related groups; Medical record homepage; Inpatient′s costs; Decision tree

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.

THE POSSIBILITY OF USING DATA MINING ALGORITHMS IN PREDICTION OF LIVE BODY WEIGHTS OF SMALL RUMINANTS

THE POSSIBILITY OF USING DATA MINING ALGORITHMS IN PREDICTION OF LIVE BODY WEIGHTS OF SMALL RUMINANTS

청각장애인의 취업결정요인 분석 연구 -데이터마이닝 기법(Exhaustive CHAID)의 적용

본 연구의 목적은 청각장애인의 취업결정요인을 데이터마이닝 기법을 적용하여 분석 제공함으로서, 장애인의 취업 성공률을 높임과 동시에 직업재활 개입의 효율성을 극대화할 수 있는 방안을 제시하는데 있다. 자료 분석을 위해 2013년 장애인고용패널조사의 제6차년도 자료를 이용하여, 전체 패널 데이터 중 청각장애인이면서 전체연령 20세 이상 65세 미만의 422명을 의사결정나무 기법의 하나인 Exhaustive CHAID 알고리즘을 적용하여 분석하였다. 본 연구를 통해서 얻어진 주요한 사실의 하나는 국민기초생활수급여부, 일상생활 도움필요 여부, 그리고 자격증 고용서비스 요인간의 상호작용(interaction)에 관한 패턴 분석이 청각장애인의 취업 예측에 주요한 역할을 할 수 있다는 것으로, 향후 직업재활 개입의 효과성을 높이기 위해 효과적인 취업결정요인, 즉 높은 학력 수준, 자격증 보유, 높은 일상생활 독립성을 가지고 있는 장애인을 적극적으로 발굴하여 집중적인 재활 서비스를 제공할 필요가 있을 것으로 사료된다. The purpose of this study was to examine demographic, human capital and service factors affecting employment outcomes of people with hearing impairments. The total of 422 individuals (age from 20 years to 65 years) with hearing impairments were collected from the Panel Survey of Employment for the Disabled from Korea Employment Agency for the Disabled. The dependent variable is employment outcomes. The predictor variables include a set of personal history, human capital and rehabilitation service variables. The chi-squared automatic interaction detector (CHAID) analysis revealed that the status of the national basic livelihood security played a determining role in predicting the employment of people with hearing impairments. Also, it was found that the three factors of the status on the national basic livelihood security, needed help about activities of dailey living, licenses & employment service factors created bigger synergy effect when they inter-complemented one another.

Effects of multifaceted consumption on happiness in life: a case study in Japan based on an integrated approach

Motivated by the ever-increasing importance of happiness in public policy and limited studies about the influences of various consumption choices on happiness, we made an initial attempt to clarify the influences of a larger set of consumption variables (77 variables covering eight major life domains) on whole-of-life happiness and on happiness at the domain level, by collecting data from 2,178 respondents residing in various cities across Japan in 2010. We proposed an integrated approach to identify factors that influence happiness based on an exhaustive Chi-squared automatic interaction detector (CHAID) approach and to quantify the degree of influence of each factor based on a Bayesian belief network (BBN) approach. The results from the exhaustive CHAID approach provide inputs to the BBN approach. As a result, 41 out of the 77 consumption variables were found to influence happiness. Whole-of-life happiness is influenced by a different set of consumption variables to happiness at the domain level. Income only influences whole-of-life happiness, but it is not the most influential factor. Saving is most important to enhancing people’s whole-of-life happiness. One’s current work-life balance does not matter for happiness. Education-related consumption variables are only associated with negative affective experiences, and surprisingly, they are unrelated to whole-of-life happiness. The effects on happiness of expenditure- and residence-related consumption variables and of consumption choices to maintain an active lifestyle are mixed. Unfortunately, none of the residence-related variables influences whole-of-life happiness. Communication with neighbors is important to positive affective experiences. Finally, we argue that happiness issues might be better captured within the framework of the life-oriented approach.

Development and Validation of Metabolic Syndrome Prediction and Classification-Pathways using Decision Trees

Purpose: The purpose of the current investigation was to create, compare, and validate sex-specific decision tree models to classify metabolic syndrome. Methods: Sex-specific Chi-Squared Automatic Interaction Detection, Exhaustive Chi-Squared Automatic Interaction Detection, and Classification and Regression Tree algorithms were run in duplicate using metabolic syndrome classification criteria, subject characteristics, and cardiovascular predictor variable from the National Health and Nutrition Examination Survey cohort data. Data from 1999-2012 were used (n=10,639; 1999-2010 cohorts for model creation and 2011-2012 cohort for model validation). Metabolic Syndrome was classified as the presence of 3 of 5 American Heart Association National Heart Lung and Blood Institute Metabolic Syndrome classification criteria. The first run was made with all predictor variables and the second run was made excluding metabolic syndrome classification predictor variables. Given that the included decision tree algorithms are non-parametric procedures, all decision tree models were compared to a logistic regression based model to provide a parametric comparison. Results: The Classification and Regression Tree algorithm outperformed all other decision tree models and logistic regression with a specificity of 0.908 and 0.952, sensitivity of 0.896 and 0.848, and misclassification error of 0.096 and 0.080 for males and females, respectively. Only one predictor variable outside of the metabolic syndrome classification reached significance in the female model (age). All metabolic syndrome classification predictor variables reached significance in the male model. Waist circumference did not reach significance in the female model. Within each model, 5 female and 3 male pathways built off of <3 American Heart Association National Heart Lung and Blood Institute Metabolic Syndrome classification criteria resulted in an increased likelihood of presenting Metabolic Syndrome. Conclusion: The proposed pathways show promise over other current metabolic syndrome classification models in identifying Metabolic Syndrome with <3 predictor variables, before current classification criteria.

The Development and Validation of an Actuarial Risk Assessment Tool for the Prediction of First-Time Offending.

For prevention purposes, it is important that police officers can estimate the risk for delinquency among juveniles who were involved in a criminal offense, but not in the role of a suspect. In the present study, the Youth Actuarial Risk Assessment Tool for First-Time Offending (Y-ARAT-FO) was developed based solely on police records with the aim to enable Dutch police officers to predict the risk for first-time offending. For the construction of this initial screening instrument, an Exhaustive Chi-squared Automatic Interaction Detector (Exhaustive CHAID) analysis was performed on a data set that was retrieved from the Dutch police system. The Y-ARAT-FO was developed on a sample of 1,368 juveniles and validated on a different sample of 886 juveniles showing moderate predictive accuracy in the validation sample (area under the receiver operating characteristic curve [AUC] = .728). The predictive accuracy of the Y-ARAT-FO was considered sufficient to justify its use as an initial screening instrument by the Dutch police.

The Effect of Additional Training on Motor Outcomes at Discharge from Recovery Phase Rehabilitation Wards: A Survey from Multi-Center Stroke Data Bank in Japan

ObjectivesThe purpose of the present study was to examine the potential benefits of additional training in patients admitted to recovery phase rehabilitation ward using the data bank of post-stroke patient registry.Subjects and MethodsSubjects were 2507 inpatients admitted to recovery phase rehabilitation wards between November 2004 and November 2010. Participants were retrospectively divided into four groups based upon chart review; patients who received no additional rehabilitation, patients who were added with self-initiated off hours training, patients who were added with off hours training by ward staff, patients who received both self-initiated training and training by ward staff. Parameters for assessing outcomes included length of stay, motor/cognitive subscales of functional independent measures (FIM) and motor benefit of FIM calculated by subtracting the score at admission from that at discharge.ResultsParticipants were stratified into three groups depending on the motor FIM at admission (≦28, 29∼56, 57≦) for comparison. Regarding outcome variables, significant inter-group differences were observed in all items examined within the subgroup who scored 28 or less and between 29 and 56. Meanwhile no such trends were observed in the group who scored 57 or more compared with those who scored less. In a decision tree created based upon Exhaustive Chi-squared Automatic Interaction Detection method, variables chosen were the motor FIM at admission (the first node) additional training (the second node), the cognitive FIM at admission(the third node).ConclusionsOverall the results suggest that additional training can compensate for the shortage of regular rehabilitation implemented in recovery phase rehabilitation ward, thus may contribute to improved outcomes assessed by motor FIM at discharge.

Modeling and Testing Landslide Hazard Using Decision Tree

This paper proposes a decision tree model for specifying the importance of 21 factors causing the landslides in a wide area of Penang Island, Malaysia. These factors are vegetation cover, distance from the fault line, slope angle, cross curvature, slope aspect, distance from road, geology, diagonal length, longitude curvature, rugosity, plan curvature, elevation, rain perception, soil texture, surface area, distance from drainage, roughness, land cover, general curvature, tangent curvature, and profile curvature. Decision tree models are used for prediction, classification, and factors importance and are usually represented by an easy to interpret tree like structure. Four models were created using Chi-square Automatic Interaction Detector (CHAID), Exhaustive CHAID, Classification and Regression Tree (CRT), and Quick-Unbiased-Efficient Statistical Tree (QUEST). Twenty-one factors were extracted using digital elevation models (DEMs) and then used as input variables for the models. A data set of 137570 samples was selected for each variable in the analysis, where 68786 samples represent landslides and 68786 samples represent no landslides. 10-fold cross-validation was employed for testing the models. The highest accuracy was achieved using Exhaustive CHAID (82.0%) compared to CHAID (81.9%), CRT (75.6%), and QUEST (74.0%) model. Across the four models, five factors were identified as most important factors which are slope angle, distance from drainage, surface area, slope aspect, and cross curvature.

의사결정나무 모형을 이용한 주관적 음성장애 예측모형

Abstract The purpose of this study was to analyze the risk factors of self-reported voice problem. Data were from the Korea National Health and Nutritional Examination Survey 2008. Subjects were 3,600 persons (1,501 men, 2,099 women) aged 19 years and older. A prediction model was developed by the use of a exhaustive CHAID (Chi Squared Automatic Interaction Detection) algorism of decision tree model. In the decision tree analysis, pain and discomfort during the last 2 weeks, age, the longest occupation and thyroid disorders was significantly associated with self-reported voice problem. The findings of associated factors suggest potential ways of targeting counseling and prevention efforts to control self-reported voice problem. Key Words : Decision tree, National survey, Prediction model, Risk factor, Self-reported voice problem * Corresponding Author : DrSc. Haewon, Byeon(Nambu Univ.)Tel: +82-62-970-0227 email: byeon@nambu.ac.krReceived April 30, 2013 Revised June 4, 2013 Accepted July 11, 2013

Where and when? Examining spatiotemporal aspects of sexual assault events

This research investigates the where and when of sexual assaults to determine what types of sexual crimes are committed at different time and place classifications. Exhaustive CHAID (chi-squared automatic interaction detection) analyses are conducted, examining factors associated with crimes that were committed indoors or outdoors, in private or public places, during the day or at night, and during the week or on the weekend. These methods are applied to a sample of 361 sexual crime events committed by 72 serial sex offenders. The results are strikingly different dependent on which spatial or temporal aspect of the crime is examined, which implies the complexity of sexual crime events and their situational components. This research brings to light possible policy implications with respect to situational crime prevention.

Soft computing techniques in ensemble precipitation nowcast

This study presented various soft computing techniques for forecasting the hourly precipitations during tropical cyclones. The purpose of the current study is to present a concise and synthesized documentation of the current level of skill of various models at precipitation forecasts. The techniques involve artificial neural networks (ANN) comprising the multilayer perceptron (MLP) with five training methods (denoted as ANN-1, ANN-2, ANN-3, ANN-4, and ANN-5), and decision trees including classification and regression tree (CART), Chi-squared automatic interaction detector (CHAID), and exhaustive CHAID (E-CHAID). The developed models were applied to the Shihmen Reservoir Watershed in Taiwan. The traditional statistical models including multiple linear regressions (MLR), and climatology average model (CLIM) were selected as the benchmarks and compared with these machine learning. A total of 157 typhoons affecting the watershed were collected. The measures used include numerical statistics and categorical statistics. The RMSE criterion was employed to assess the suitable scenario, while the categorical scores, bias, POD, FAR, HK, and ETS were based on the rain contingency table. Consequently, this study found that ANN and decision trees provide better prediction compared to traditional statistical models according to the various average skill scores.

Comparison of multilabel classification models to forecast project dispute resolutions

Early forecasting of project dispute resolutions (PDRs) provides decision-support information for resolving potential procurement problems before a dispute occurs. This study compares the performances of classification and ensemble models for predicting dispute handling methods in public–private partnership (PPP) projects. Model analyses use machine learners (i.e., Support Vector Machines (SVMs), Artificial Neural Networks (ANNs), and Tree-augmented Naïve (TAN) Bayesian), classification and regression-based techniques (i.e., Classification and Regression Tree (CART), Quick, Unbiased and Efficient Statistical Tree (QUEST), Exhaustive Chi-squared Automatic Interaction Detection (Exhaustive CHAID), and C5.0), and combinations of these techniques that performed best for a set of PPP data. Analytical results exhibit that the combined technique of QUEST+CHAID+C5.0 has the best classification accuracy at 84.65% in predicting dispute resolution outcomes (i.e., mediation, arbitration, litigation, negotiation, administrative appeals or no dispute occurred). Moreover, as the dispute category and phase in which the dispute occurs are known during project execution, the best classification model is the CART model, with an accuracy of 69.05%. This study demonstrates effective classification application for early PDR prediction related to public infrastructure projects.