Numerical prediction method of BP neural network based on random forest feature extraction

Abstract

The invention relates to a numerical prediction method of a BP neural network based on random forest feature extraction, and belongs to the field of machine learning and data mining, and is characterized by comprising the following steps of: (1) selecting out-of-bag data X1 (x1, x2, and the like, xn), and calculating out-of-bag data errors error (1); (2) adding noise interference to the characteristics T (t1, t2, and the like, tm) of all the samples of the out-of-bag data randomly, and calculating out-of-bag data errors error error error error (2) again; (3) calculating the importance I of allthe features T (t1, t2, and the like, tm); (4) determining a characteristic T (t1, t2, and the like, ts) with relatively high importance and the importance degree (1, 2, and the like, s) of the characteristic T; (5) training the BP neural network, and updating the connection weight i; And (6) taking the test set samples as input, carrying out feature selection, and inputting the selected featuresinto a BP neural network for prediction to obtain a prediction result. According to the established numerical prediction method of the BP neural network based on random forest feature extraction, thefeatures of the sample set are extracted through the random forest, the importance degree of the features is quantified, and numerical prediction is achieved. It can be known through multiple sets ofdata experiment comparison results that the prediction method is numerical prediction, and the prediction method for enhancing the generalization ability of the model on the basis that the predictionprecision is guaranteed is provided.

Description

technical field [0001] The invention relates to the field of machine learning and data mining, and mainly relates to a numerical prediction method. Background technique [0002] At present, for numerical prediction problems, most models can fit the original data to a high degree, but the model generalization ability is poor. These models tend to show good predictive performance on the training set, however, on the test set or some emerging data, the predictive power is greatly reduced. Although the neural network has a good generalization ability, the convergence speed is too slow during the model training process, resulting in a long training time, which cannot meet the timeliness requirements of numerical prediction. Taking BP neural network as an example, the early BP neural network has problems such as slow convergence speed and easy to fall into local optimum. Although the BP neural network uses the method of error back propagation to adjust the weights between neuron...

AI Technical Summary

Problems solved by technology

Although the BP neural network uses the method of error backpropagation to adjust the weights between neurons in each layer, but in the initial modeling, the connection weights between the input layer neurons and the first hidden layer neurons It is a random number set by the initialization method, which may be a reason for subsequent network training to fall into local optimum

Images