|
We use the random forest in the concept of decision tree to discuss the binary classification problem (Classification) with imbalanced data (Imbalanced Data). The proportion is much larger than other types of data, and it can also be explained in reverse. One of the data is significantly lower than the other data. When the evaluation criteria cannot accurately understand the situation of the data, it may lead to misjudgment in decision-making or failure to discover the truth. and so on. In such a situation with imbalanced data characteristics, if events are to be predicted to belong to a critical minority class, it is mainly necessary to find out why the lower error value occurs, and the event rate is less than 5%, then by understanding the process to Improves the probability of success in judgment and decision-making. Therefore, it is not suitable to use the traditional evaluation method to measure the performance of the model correctly, that is, it is not suitable to use the accuracy of everyone's perception as a measure to judge the quality of the model. We must find out the reasons for the imbalanced data to improve the accuracy. This study will take the case of a patient in a surgical ward of a hospital in southern China as an actual case. First, after excluding cases of extubation at the end of life and cases with incomplete data, the remaining data is the patient's first extubation, with a total of 2,783 One patient was extubated for the first time, and 174 of them needed to be re-intubated, and about 6.25% of them failed extubation. The hospital hopes to use the extubation data of these 2,783 patients to find out the important factors that lead to reintubation, or what characteristics are there, and to further establish a feasible prediction model, hoping to help doctors in extubation. It can be predicted that which type of patients may be a high-risk group for extubation failure, which conditions may lead to extubation failure, which type of patients can have a high probability of successful extubation, and which conditions may have a high probability of successful extubation. Despite the evaluation of metrics of success and failure, this is the motivation for this study to explore binary classification.
|