Make sure that both are related to same problem, like Action Recognition or Text classification etc. Generally, as you mentioned, we use the dataset and split it into two sets.

That's right. You can also change the aspect ratio and affect the accuracy of the model.

It would also be even better if you still split into a validation set, for intermediate validation of the network training process.

Yes, you can do that provided these datasets have the same number of features.

both datasets have the same column name and datatype.

Dear Riken Mangukiya,

The split is being done to validate the performance of your designed model before deploying it; as you want to bypass this step, go for the testing directly with the test dataset you have, and no need to split your data. My advice is to validate the performance of your system first.

Thanks all for sharing your knowledge and experience with me on this topic.

Hello sir, @Mohamed Elhadad

I have followed exact same thing before using a separate testing data set. But I did not get that much accurate prediction algorithm model as I am getting in separate testing data set.That's why I wanted to know if am I doing right or not.

Thanks for your time and valuable answer.

Akhtar Jamil Akpofure A Enughwure Shahboz Sharifovich Qodirov Mohamed Elhadad Jakob Laue

First of all, thank you for giving a reply to this question.

but still, I have one follow-up question regarding this, could you explain it, please.

It is about validation and selecting the best model for the dataset.

As per the possibility of using two different datasets in one classification model.

How could I use the K-cross validation model? as per my opinion, I think K cross validation possible when we are using one dataset and splitting it into two different data set for training and testing, right?

Can I use the Grid Search parameters method to validate my model? and can I Compare the ML model's accuracy or f1_weighted score to selecting the optimum model?

OR

Is it valid to just compare model accuracy from the Confusion matrix and choosing the best model for prediction?

I will be very grateful to accept all other's answers and still want to learn more in this field.

Thank you all!

Dear Riken Mangukiya ,

The accuracy, sometimes, might give you a misleading evaluation, e.g., if the dataset is unbalanced. I recommend that you use Geometric Mean (GM), or Area Under a Curve (AUC) for evaluation.

regarding the use of two datasets, you must choose the classifier that gives you the highest results on both datasets. Then, you will try to optimize your, models, playing with hyperparameters, to get the best performance from your classifier. it should be noted that when optimizing your classifier you have to avoid the underfitting, overfitting problems; your goal should be to develop a generic system!.

I hope that helps!!

Regards

Mohamed Elhadad Sir, Again thank you for giving a recommendation and help. I will try with your idea and make sure my model will be accurately prepared.