The Application Study Of Random Forest Altgrothim Predicting Patients With Hospital Acquired Infection

With the rapid development of health care information, more and more medical problems can be solved by computer technology, medical problems get more attention. The hospital infectious disease is the second leading cause of death, hospital acquired infections will not only bring physical and financial burden of patients, can also cause an unreasonable use of hospital resources. Now the methods for monitoring and predicting of hospital acquired infections mainly are ex post judgment methods, there is a certain lag, however advance monitoring tools showed a single, incomplete features, the problems of hospital acquired infections prediction is immediately to be solved. In this paper, based on improved random forest model predicts the risk of infection in hospitals, makes a new attempt for the rapid prediction of hospital acquired infections risk and improve prediction accuracy.First, analyzing the characteristics of patients with nosocomial infection data, according to the input requirements of these characteristics and the random forest algorithms for pre-processing raw data set, in order to obtain the most comprehensive, high-quality feature set.Second, there is an imbalance problem for medical raw data. The infected data is far less than the non-infected infection characteristic data. This paper uses SMOTE algorithm to balance unbalanced data, and improves classification accuracy of model predictions.Third, this paper proposes an advanced scalable random forests algorithm to give an accurate prediction of whether a new sample of nosocomial infections. Improved extended random forests algorithm is composed of some random forests, each of those is generated by different infected type data, respectively. Otherwords the scalable random forests are combined with random forest, and then applies hadoop component and mahout machine learning library to expand random forest algorithm.Finally,this paper analyses the First Hospital of Qinhuangdao 14223596 sets of clinical data, extracts experiment data and then equilibrated with clinical data, the use of improved random forest algorithm for training and testing, and then use the index to evaluate the classification algorithm to predict the results, and compared with traditional random forest algorithm predicting results, there shows that the improved algorithm prediction accuracy in medical clinical data were significantly better than traditional random forest algorithm.