Forecasting with ANFIS and SVR

In MATLAB, is it possible to make forecasting of unknown values using ANFIS or SVR? For example, I have a time series from 2010 to 2020, and I would like to make forecasting for the years 2021, 2022, and 2023. I have already tried using the MATLAB's toolboxes and in line of code, but I have not been successful.

Answer 1

[Anay]

Hi Vandilson,

I understand that you want to perform timeseries forecasting using ANFIS or SVR in MATLAB. It is possible to do timeseries forecasting using ANFIS. In the MATLAB documentation, you can find the example “Predict Chaotic Time Series” which demonstrates how to predict a time series with ANFIS. You can follow the below link to view the example:

https://www.mathworks.com/help/fuzzy/predict-chaotic-time-series-code.html

You can utilize SVR as well for time series forecasting. You need to put the training data in proper format and then train the SVR model using the Statistics and Machine Learning Toolbox of MATLAB. You can consider following these basic steps to process your data:

1. Extract relevant features from the data, lets call them X
2. Ensure X is in proper time series format. Split X into train and test data sets
3. Decide a timestep “t” for your model. Your model will be trained to take previous (t-1) points in input and predict the output at timestep “t”
4. Put your train and test data into format of the time step. It means that each row of the data will have “t” columns: first (t-1) columns as input variables and the last column as target variable

I hope this helps resolve the doubt!