In my humble experience, Random Forest, because it tries regression with many trees choosing the best possible this gives a chance to optimize the result giving a great advantage over other algorithms such as a simple tree,, and mixing two algorithms that in my experience work very well together,

First, thanks Mariana to the interest. I had previously used random forests, but its dependence on two types of randomness in their work, as you know, therefore it's succeeded in the training process, but it's hard to succeed in the testing process because the concept of randomness. What's your opinion on Tree Net is also one of the Data mining algorithm, I hope to hear from you soon.

With best regards.

samaher

I doubted the performance of CHAID and E-CHAID. When my son asked me the thema of his graduation work, I suggested the comparison of SPSS four methods. Above both methods are inferior to CART. I guess multi-branch in the upper levels cause bad results. However, this results are valid for his data. Not general.

Thank you Shuichi Shinmura very much for your'i interesting, but i'm suggest the

1. A Novel Agent-DKGBM Predictor for Business Intelligence and Analytics toward Enterprise Data Discovery ,

Today’s business environment requires its workers to be skilled and knowledgeable in more than one area to compete. Data scientists are expected to be polyglots who understand math, code and can speak the language of business. This paper aims to develop a new Agent- Develop Kernel Gradient Boosting Machine (Agent- DKGBM) algorithm for prediction in huge and complex business databases. The Agent-DKGBM algorithm executes in two phases. In the first phase, building the cognitive agent, the primary goal is to prepare the database for the second phase, searching the business databases. During this phase, the cognitive agent selects one of the business databases, choosing the most suitable type i.e., Hyperbolic functions, Polynomial functions and Gaussian mixture as a kernel of Develop Support Vector Regression (DSVR) and determines the optimum parameters of the DSVR and DKGBM. The second phase consists of three stages, which include splitting the business databases into training and testing datasets by using 10-fold cross validation. In the second stage, a DKGBM model using the training data set is built to replace the Gradient Boosting Machine (GBM) kernel, typically using Decision Trees (DTs) to produce the predictor with DSVR because it would potentially increase the accuracy and reduce the execution time in the DKGBM model. Finally, the DKGBM would be verified based on the testing data set. Experimental results indicate that the proposed Agent-DKGBM algorithm will provide effective prediction with a significant high level of accuracy and compression ratio of execution time compared to other prediction techniques including CART, MARS, Random Forest, Tree Net, GBM and SVM.

The results also reveal that by using Gaussian mixture as a kernel of DSVR, the Agent-DKGBM achieves more accurate and better prediction results than other kernel functions, which prediction algorithms typically use, also

than GBM, which typically use DTs. Results clearly show that the proposed Agent DKGBM improves the predictive accuracy, speed and cost of prediction. In addition, the results prove that Agent-DKGBM can serve as

a promising choice for current prediction techniques.

2. Modern Predictor Model for Healthcare problem (MPM-HCM).

level of health of society in general. This management requires providing prospective information about the need for patients admitting in a hospital, and the necessary medical resources. Prediction techniques represent an effective tool for knowledge discovery in huge and complex datasets in many fields including healthcare. Methods: In this work, we design and implement a prediction model called a Modern Prediction Model for HealthCare Problem (MPM-HCP) which introduces two improvements for Gradient Boosting Machine (GBM) prediction technique. MPM-HCP developed (GBM) by inspiring positive sides of linear regression to replace splitting criterion with a correlation measure in regression tree building. It also reduced the complexity of building boosted model by using a fast method for choosing best split point. Results: MPM-HCP has significant behavior in terms of prediction error and execution time. In comparison with tradition gradient boosting trees, the MPM-HCP has a testing error of 0.468, while original GBM based on sum of squared has error of 0.491, and original GBM based on standard deviations has 0.481 error. Training time is also reduced more than 85%. Conclusions: MPM-HCP implementation showed that there were three attributes frequent in binary regression trees building. Those attributes were gender of patient, number of claims to admit hospital, and the medical procedure group, which means those attributes is more correlated with the target of prediction (i.e. number of hospitalization days). MPM-HCP confirms the ability to produce precise prediction result, and the scalability to deal with huge dataset in suitable execution time.

Both papers based on develop of  GBM by two different methods that give results best  compare with other methods. therefore, i'm advice of used it. 

Thank you for your kind explain.

I left the decision tree  now.. So, I just offer my son's results as your reference.

I adviced  long-term care insurance system in Japan. My friend, medical doctor, developed it by CHAID. At first, it had many defects. Now, it was rebuilt by CART.

Now, I engage to build new discriminant analysis, because I developed the diagnostic logic of an ECG by Fisher's LDF and QDF. But it was inferior to the decision tree developed by the doctor. I almost complete my research after 2014. I developed the "100-fold cross bvalidation" method. At first, I tried 10-fold cross validation, but I cannot obtain the 95% CI of error rate and discriminant coefficients. I recommend you change 100-fold CV.