Development of a models to predict colorectal cancer warning sign

Abstract

This research aims to study the most suitable predictive model and to develop a risk assessment model to be used as a warning sign for colorectal cancer. There are 5 classification algorithms, namely Naive Bayes Classifier, Decision Tree, Random Forest, Deep Leaning and Logistic Regression Analysis. By comparing the prediction performance, Accuracy, Precision, Recall, and Mean Square Error (RMSE). This study collected data of 396 patients in Nakhonpathom Hospital with 7 attributes. In the research method, the data was divided into 5 parts or 5 Fold Cross-Validation by randomly dividing the data into 5 parts and selecting 4 parts of data called training data to create a model for both 5 algorithms and randomly selecting 1 part of data called validation data (Testing Data) is used to assess the error of the model. The results of the research showed that the Logistic Regression Analysis method, the accuracy, the precision, the recall and the RMSE were 92.41% 0.90 1.0 and 0.24, respectively, because it is a predictive technique, it will be used when the dependent variable is a quality variable. As for the independent variables or predictive variables, they can be both qualitative (sex) and quantitative variables with continuous values (age and Carcinoembryonic Antigen (CEA), Haemoglobin (HGB), White Blood Cells (WBC), Haematocrit ( Hct), Thrombocyte (Plt). Therefore, Logistic Regression Analysis method has the best predictive efficiency.