Prediction method for time sequence intervention effect

Abstract

The invention discloses a method for predicting a time sequence intervention effect. The method comprises the following steps: acquiring a data set; constructing a deep learning model; enabling two depth representations formed by the same input in the two recurrent neural network units to form a pair, and applying orthogonalization constraints to the pair; defining a total loss function, and minimizing the total loss function to realize training optimization of the model; training on the training set by adopting different hyper-parameter combinations, verifying the performance on the verification set, selecting the optimal hyper-parameter combination, and testing the deep learning model trained by adopting the optimal hyper-parameter on the test set to obtain a predicted intervention effect. According to the prediction method provided by the invention, related representations of intervention measures and intervention effects are respectively learned by utilizing two recurrent neural network units through a multi-task learning strategy, hybrid factors causing selection bias errors are eliminated through orthogonalization constraints, and finally, decoupling representations on a time sequence are obtained through training and are used for predicting the intervention effects of the time sequence.

Description

Technical field [0001] The present invention belongs to the field of data processing, and specifically, the prediction method of timing intervention effects. Background technique [0002] Clinical decision makers often face problems to choose interventions for patients, and intervention effect is the basis for decision making. Therefore, it is important to estimate its effect. Although clinical random controlled experiments represent causal inferred gold standards, they have high cost, and there are fewer patients. Compared to the real world clinical data represented by electronic medical records has the characteristics of low cost, large amount of data, large coverage, and uses these data to conduct an increasingly wide concern. However, using real world clinical data to conduct intervention effect assessments will face two major challenges: anti-fact results lack and choose a bias. Since a patient can only accept one intervention program at the same time, we have no results of ...

AI Technical Summary

Problems solved by technology

However, the counterfactual results in time series and the formation mechanism of selection bias are more complicated

Images