Search

This research aims to study the performance of data mining techniques in medical datasets. The data in this research contaita information of patients with breast cancer, diabetics and patients with hyperthyroidism. All datasets were collected from UCI databasee Mmachine learning, in particular Decision Tree C4.5, Naïve Bayes, Neural Networks, Random Forest and Deep Learning techniques were used to create the models of disease Breast cancer, diabetes and hypothyroidism prediction models. In order to measure the performance of prediction models, 10-fold cross validation was utilized to divide the data into training and testing sets. Accuracy, sensitivity and specificity of the prediction models were used to compare the prediction performance of each model. The experimental results showed that the Decision Tree C4.5 technique was the best technique in modeling the prognosis of hypothyroidism. It provided 99.86 % accuracy, 99.85 % sensitivity and 100 % specificity. [ABSTRACT FROM AUTHOR]

This research aimed (1) to create cardiovascular risk diagnosis prediction models using algorithms including Neural Network, Random Forest, Naïve Bayes, K-Nearest Neighbors and Decision Tree (2) five algorithms were used with attribute selection and (3) comparing the model performance using 10f Fold cross validation method. Tools usee were MySQL and RapidMiner Studio programs. The data see comprised people who had been screened as patients with cardiovascular disease that were collected from the Saraburi Provincial Public Health Office during 2018-2019 from 12 Saraburi hospitals and 126 health promoting hospitalsreIt was found that the model with the best prediction performance was the neural network model with attribute selection having 99.29% accuracy, and the lowest was the decision tree model with 70.39% accuracy. This research concluded that the neural network model with attribute selection of the best qualificationt should be further developed for early diagnosis of cardiovascular risk web applications. [ABSTRACT FROM AUTHOR]

Text mining is one of the most effective data analysis processes using alphabetic methods. Currently, text mining techniques are classified in a variety of ways. This research aims to find the most effective of 5 techniques that were Naïve Bayes, Support Vector Machine (SVM), K-Nearest Neighbor C4.5, and Random Forest. The data collected, in total of 3,798 messages, were all made by the viewers. The categorization process divided the data into 2 groups: positive character and negative character. Interestingly, the process has only indicated selection of adverbs and slangs as a core division to produce positive and negative characters. After analyzing the data, two problems were found class imbalanced. SMOTE were used for filtering and to increase the minority class. 10-fold cross validation was applied to segment the data into training and testing sets. Moreover, precision recall and accuracy are used as the criteria for selecting the most effective model. The results showed that the K-Nearest Neighbor produced greatest accuracy in categorizing the messages with a precision of 99.75% recall of 100% and accuracy score of 99.87%. [ABSTRACT FROM AUTHOR]

เท 2018, The World Health Organization (WHO) and Department of Mental Health (DMH), specified that major depressive disorder (MDD) was the second most important disease that it is probably caused by social media usage affecting stress and leading to violence, and depression . This research proposes the depressive classification from posts on twitter of user behaviors and compared the accuracy of two classifiers between one level and two levels:- (1) one level: using the Bayes algorithm created a model for classification between general and symptoms based on a symptoms detailed in a questionnaire (DSM-5) including as follows: depression, loss of interest, loss of appetite, abnormal sleep, slowed thinking, guilt, tiredness, unexplained and suicidal ideation. (2) Two levels: Using the SVM algorithm created a model for classification between general and depression. Using the Bayes algorithm compared with the Random Forest algorithm for classification of symptoms เท a questionnaire (DSM-5). The data came from real postings of international celebrities. The dataset is divided into 2 sets: a training set and a test set. Finally, the results are demonstrated in a training set prediction between one level and two levels: One level: the Bayes algorithm showed that the accuracy=82.55%, and the SVM algorithm showed that the accuracy=96.18%. Two level: the SVM algorithm showed that the accuracy=98.20%. SVM algorithm pair with Bayes algorithm showed that the accuracy=82.23%, and SVM algorithm pair with the Random Forest algorithm showed that the accuracy=91.45%. The results of test set, by the boundary of probability are variously set 0.1 to 0.9 that prediction between one level and two levels : One level: the Bayes algorithm showed that the accuracy=76.67%, and the SVM algorithm showed that the accuracy=70.00%. Two level: SVM algorithm pair with Bayes algorithm showed that the accuracy=73.33%. SVM algorithm pair with Random Forest algorithm showed that the accuracy=70.00%. [ABSTRACT FROM AUTHOR]