how to use predict and forecast

2 次查看（过去 30 天）

DARLINGTON ETAJE 2019-10-21

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/486586-how-to-use-predict-and-forecast

评论： DARLINGTON ETAJE 2019-10-21

Hello Family,

I am trying to predict the class of future dataset: Unfortunately my code is completely misclassifying one class meaning there is a problem with the code.

This is the dataset:

https://drive.google.com/file/d/1tQdZ5XXQAECN6eahzhkt5xLpZA9eOyFk/view?usp=sharing

Here is the code I am currently using:

%D be the dataset
Data=D(1:500,1:end-1);
Data2=D(501:1000,1:end-1);
class_labels=D(1:500,6);
Data2labels=D(501:1000,6);
classification_model = fitcensemble(Data,'class_labels~var1+var2+var3+var4+var5')
cv = cvpartition(classification_model.NumObservations, 'HoldOut', 0.2);
cross_validated_model = crossval(classification_model,'cvpartition',cv);
Predictions_1 = predict(cross_validated_model.Trained{1},Data(test(cv),1:end-1));
Predictions_5 = predict(cross_validated_model{1}.Trained,Data2);
training_data = Data(training(cv),:);
labels = unique(class_labels);
Y = ismember(training_data.class_labels,labels{1});
Results_1 = confusionmat(cross_validated_model.Y(test(cv)),Predictions_1);
Results_5 = confusionmat(Data2labels,Predictions_5);

2 个评论
显示无隐藏无

John D'Errico 2019-10-21

It does not mean there is a problem with the code. There could easily be a problem with the data, such that this particular set of data will never be able to predict some events correctly, no matter what model you use. It is not even that difficult to create such a dataset.

DARLINGTON ETAJE 2019-10-21

Hello John,

Thanks for your comment. That is not the case here. There is a 95 percent accuracy when using the dataset as one..meaning using holdout to divide the data into training set and testing set. But when predicting the future, the story changes....I am positive it's a problem with the code...

请先登录，再进行评论。

请先登录，再回答此问题。