Accuracy Estimation Based On The Difference Between Test Samples And Train Samples

Classifier is an important part of machine learning, there are many applications beused in real life. The classification accuracy is an important standard to measure thequality of the classifier. A classifier give classification results without accuracy rate.Accuracy rate is assessed later. Problems such as classification accuracy drop cannotbe timely and effectively found. It is necessary that mark all test samples when gatheraccuracy rate and it is expensive to mark all test samples.Because classifier applied more and more widely, classification accuracyestimation is becoming more and more important. A high accuracy rate is favorable.We hope to make sure a high accuracy rate, at least make sure accuracy rate can keepin the training accuracy level. The classification accuracy tends to drop slowly as timegoes by. It is meaningful to discover accuracy drops in time.It is a problem that we care about that how much is the classification accuracyrate when a group of new samples are gathered. To solve this problem, this paper putforward different estimates from different perspectives. One estimate is based on thedifference in sample distributions. Another estimate is based on the difference in eachsample. Classification accuracy is estimated by using a function approximation method.The estimation of classification accuracy is applied to citrus canker recognition systemfinally.