How can I improve the accuracy of classification of real-time time series data?

Although accuracy is high when learning time series data (AC voltage or current value) with a cycle on a 1-D CNN,

In fact, if input data is input one at a time (because it is actually input like that), the accuracy is very poor. Is there a way to improve accuracy?

(Except for data preprocessing techniques...)

Answer 1

[Yash]

Hi Jiwon,

I understand you want to improve the accuracy of classification of real-time time series data. One way to improve the accuracy of a 1-D CNN for time series data is to use a sliding window approach. Instead of feeding the data one sample at a time, you can group several samples together and feed them to the network as a batch. This can help the network capture temporal dependencies and improve its accuracy.

To know how to implement sliding window algorithm in MATLAB, you refer to the following accepted MATLAB Answer: https://in.mathworks.com/matlabcentral/answers/24923-how-to-implement-sliding-window-algorithm-in-matlab

Hope this helps!