目的构建数据不平衡时癫痫发作频率预警模型。方法以2008年9月～2011年1月在临沂市人民医院就诊的736例癫痫患者为研究对象,进行癫痫发作危险因素的流行病学调查。采用Smote算法,进行数据的平衡化处理、后构建随机森林模型(RF),对癫痫患者的发作频率(次/月)进行分类预测。结果采用随机森林模型对癫痫患者进行分类,判断的正确率为82.53%,误判率为17.47%,受试者工作特征曲线下的面积为94.2%,袋外误差率为13.3%。结论随机森林法能够对癫痫患者发作的频率进行快速的分类,为癫痫发作的预警提供科学依据。 Objective To construct the early warning model of epileptic seizure frequency when the data is unbalanced. Methods Seventy-six patients with epilepsy from September 2008 to January 2011 in Linyi People’s Hospital were enrolled in this study. Epidemiological investigation of the risk factors of epileptic seizures was conducted. The Smote algorithm was used to balance the data. Then, a random forest model (RF) was constructed to predict the seizure frequency (time / month) of epileptic patients. Results The random forest model was used to classify the patients with epilepsy. The correct rate was 82.53% and the false positive rate was 17.47%. The area under the receiver operating characteristic curve was 94.2% and the error rate was 13.3%. Conclusion The random forest method can quickly classify the frequency of seizures in patients with epilepsy and provide a scientific basis for early warning of seizures.