Khoa Truong Dinh, Sławomir Kukliński, Tomasz Osiński, Jacek Wytrębowicz, Heuristic traffic engineering for SDN, Journal of Information and Telecommunication, 2020, DOI: 10.1080/24751839.2020.1755528, https://www.tandfonline.com/doi/full/10.1080/24751839.2020.1755528
Existing research papers mostly with smaller datasets. The best way is you can create more nodes and controllers and assign the configuration using mininet simulator
I definitely agree with Ramkumar M P . The best way is to simulate your own topology using Mininet and GNS3. Moreover, you can write your code with python and integrate it into the controller of the SDN.
You can use any dataset generated for TCP/IP-based network for traffic classification. SDN also uses TCP/IP and all the header fields will be same. It is better to use benchmark datasets for training and Mininet based SDN environment for generating test data. One of the dataset for traffic flow classification can be found at :
You can find many of the trusted datasets for SDN in any good review paper published in ISI or Scopus journals such as IEEE, Springer, ACM, MDPI, and other journals.