Sheetal, by predicting discrete values, do you mean classification (different categories) or regression (the target variable that can take discrete values only)?
of course it depend on the dimension of features and patterns in data. but, "generaly speaking", I would rank them like this: Xgboost> random forest> SVR > KNN
Teuku, it all depends on the nature of the data! for example if your data has a linear structure, then the linear kernel should suffice. However, it is not always obvious what the data structure is so in practice, you should apply multiple kernels and see which one preforms better.
Hello, thanks a lot for this lesson. I really enjoyed it! One question. What software would you suggest me to use for statistical analysis among gretl, jasp jamovi? My background is in economics, but I theoretically covered quite a lot of the machine learning topics, and some deep learning. What is the most powerful free software around that doesn't need coding? Thanks a lot
Amazing Video! I learned so much from your videos. Thank you so much for making these Videos. What's the best way to check the model performance for SVR? For example we use MSE score for linear regression. But I don't think it will apply to SVR as it uses a different loss function. I read somewhere MAE is a good option for SVR can you shed some light on this please?
Excellent point Ash. In SVR, we use "epsilon-insensitive" loss function which is (kind of) v-shape and for that matter you should use MAE. However, as you may notice at 10:15, depending on what norm we use to penalize slack variables (L1 or L2) we can use either MAE or MSE respectively.
@@pedramjahangiry for a simple SVR model from sklearn with rbf kernel, which penalty is used L1 or L2? It's quite confusing as documentation those refer to 2 loss function but doesn't clearly specify which is used when
@@Ash-bc8vw as stated in the sklearn documentation (scikit-learn.org/stable/modules/generated/sklearn.svm.SVR.html), "the penalty is a squared L2 penalty".
Sir Pedram thank you a lot for the informative video, I would like to ask you, how can I apply SVR with the panel data regression model by using R language?? your answer is to grant me a lot of help...,thank you in advance
Hasan, I don't do machine/deep learning with R but I encourage you to ask the same question from ChatGPT. there are tons of good tutorials out there. Let me know if you couldn't find one.
Sir, can you make video to explain the implementation of svr based on time data series manually? I mean not using software like python. Many tutorial I have ever seen is only explain the concept but not explain the implementation
Idris, for time series everything should be the same. for example in univariate time series, just assume y = p_t and X=P_(t-1). If you are asking for manual calculations, I haven't done it myself. But will put in in my to do list. thanks for your feedback.