Daraei Atefeh, Hamidi Hodjat
Dept. of Information Technology, Faculty of Industrial Engineering, K.N. Toosi University of Technology, Tehran, Iran.
Iran J Public Health. 2017 May;46(5):682-692.
Myocardial infarction (MI) occurs due to heart muscle death that costs like human life, which is higher than the treatment costs. This study aimed to present an MI prediction model using classification data mining methods, which consider the imbalance nature of the problem.
We enrolled 455 healthy and 295 myocardial infarction cases of visitors to Shahid Madani Specialized Hospital, Khorramabad, Iran, in 2015. Then, a hybrid feature selection method included Weight by Relief and Genetic algorithm applied on the dataset to select the best features. After selection of the features, the metacost classifier applied on the sampled dataset. Metacost made a cost sensitive J48 model by assigning different costs ratios for misclassified cases; include 1:10, 1:50, 1:100, 1:150 and 1:200.
After applying the model on the imbalanced dataset, the cost ratio 1:200 led to the best results in comparison to not using feature selection and cost sensitive model. The model achieved sensitivity, F-measure and accuracy of 86.67%, 80% and 82.67%, respectively.
Experiments on the real dataset showed that using the cost-sensitive method along with the hybrid feature selection method improved model performance. Therefore, the model considered a reliable Myocardial Infarction prediction model.
心肌梗死（MI）是由于心肌死亡导致的，其代价如同人类生命，高于治疗成本。本研究旨在使用分类数据挖掘方法呈现一个心肌梗死预测模型，该方法考虑了问题的不平衡性质。
2015年，我们纳入了伊朗霍拉马巴德市沙希德·马达尼专科医院的455名健康访客和295名心肌梗死患者。然后，一种包括基于 Relief 的权重和遗传算法的混合特征选择方法应用于数据集以选择最佳特征。在选择特征后，元成本分类器应用于采样数据集。元成本通过为误分类案例分配不同的成本比率来创建一个成本敏感的 J48 模型；包括 1:10、1:50、1:100、1:150 和 1:200。
在不平衡数据集上应用该模型后，与不使用特征选择和成本敏感模型相比，成本比率 1:200 产生了最佳结果。该模型的灵敏度、F 测度和准确率分别达到了 86.67%、80%和 82.67%。
在真实数据集上的实验表明，使用成本敏感方法和混合特征选择方法提高了模型性能。因此，该模型被认为是一个可靠的心肌梗死预测模型。
