I am getting the following results for my SVR Model using Python's Sci-kit library :
SVR RESULTS - C AND GAMMA PARAMETERS OPTIMIZED BY GRID SEARCH
- RMSE = 0.17050234208788798
- MSE = 0.029071048657455172
- Normalized RMSE= 0.17050234208788798
- R Square = 0.661724392932528
- K-fold (10 fold) Cross validation accuracy mean 0.833655796514621
Could you please let me know which one's should be valid ? r_square I calculated manually and do not think it would work for non-linear data.
Normalized RMSE was calculated as follows: nrmse=rmse/(y_test.max() - y_test.min())
Please suggest which metric would be the best to describe my model
Thanks in advance,
Bhaskar
Thank you Mohammed Deeb for your insightful answer.
I am trying to Optimize SVR hyper parameters with different meta-heuristic approaches like Particle Swarm Optimization-SVR (PSO-SVR), Genetic Algo-SVR, Grid Search-SVR, Tabu Search-SVR etc. Then, based on the optimized Cost, Gamma parameters of 'rbf' kernel based SVR, I will compare the forecasting accuracy of different SVRs to determine which hybrid approach for SVR optimization works best for me. I am using only SVR but combining with different hybrid approaches of the same family of algos.
In this case , do you recommend MAE would be good as you mentioned above ?
Regards,
Bhaskar T
Hi Mohammed Deeb - Based on some literature review, I have decided to use
all RMSE,MAE and MAPE as they give a better view of the data. I have one more question - do you think mentioning K-fold (10 fold) Cross validation accuracy mean is advisable for reporting results. I have not seen too many papers using this metric in high impact papers.
Looking forward for your advise on this.
Regards,
Bhaskar
- Badges
- Science topic
- Similar topics
- Engineering
- Modeling