Application of BP improved algorithm in the prediction of asthma symptom-syndrome classification

Summarize:

First, research content

In this paper, we study the application of CAL-BP (adaptive improved BP algorithm based on the hidden layer of competitive learning and learning rate) in the classification and prediction of symptom syndromes.

Second, the idea of arithmetic

1, after the hidden layer calculates the error of each node, the weight of the node with the maximum error is corrected normally,

and the weights of the other units are to the opposite side. to the correction, the value correction amount of the hidden layer node is expressed by Δ,

The correction amount of the adjustment formula is specific to

　　　　

2. After each iteration of the algorithm, the value of the error function is calculated and compared with the previous value, if the value of the error function increases,

Then the learning rate is adjusted, and the learning rate should be lowered at a certain rate in the next iteration, if the i+1 value of the error function is reduced,

On behalf of the learning rate increase can be increased, with Z for the first iteration of the learning rate, e+ representative + 1 iterations

The error function changes the value, the learning rate for the first +2 iterations is:

Iii. Conclusion

In this paper, the improved algorithm, the error after repeated modification can be fixed at 0.0403, the recognition rate of 83.6%, training time from the general 1 minutes 40 shortened to 11 seconds.

CAL-BP is faster than the normal BP, training time is little, the recognition rate is high.

　

Iv. My understanding

Algorithm aspect: The preprocessing of the data (input data processing and quantization processing and output data quantization processing) overall has a grasp, the algorithm used to feel more abstract place, materialized.

Thinking: Using the data at hand to do a simulation of the concrete practice of a step forward.

V. Parts that do not understand

Where the quantification is described, the value of the quantization value has a place where there is no clear ( fluorescent color: Why three kinds of syndromes, the quantization value of {0 1 2 3}), need to be re-pondered or consulted.

　　

The algorithm thought also has the place which does not understand, is the formula expression source.

Other is the need to practice their own practice. Go on, come on.

　　

Application of BP improved algorithm in the prediction of asthma symptom-syndrome classification