Erik Cuevas thanks for your suggestion, can you please tell which one is good ways for resampling of data among bootstrapping, cross-validation, repeated cross-validation, to improve accuracy or i have to try each one one by one to choose that coefficients(model) which give maximum accuracy .

When data are scare, one way is to resamplig avaliable data in order to get more elements

it depends on what you are doing. If you have classification problem (e.g. with 3 classes as outcome), then I recommend to adapt thresholds for classification.

This can be very effective.

This usually increases accuracy, sensitivity, specificity, negative or positive predictive values.

The disadvantage is that not all objects will be classified.

I use this approach in biomedicine sometimes. ...

This works also of course if you have image data ...

in case of any questions, let me know...

Wolfgang

Wolfgang Hitzl i am working on 2 types of problem first one is Regression based and second one is classification based.

itz not a image data.

Erik Cuevas thanks for your suggestion, can you please tell which one is good ways for resampling of data among bootstrapping, cross-validation, repeated cross-validation, to improve accuracy or i have to try each one one by one to choose that coefficients(model) which give maximum accuracy .

Luciano da F. Costa Thanks

refer this site

https://medium.com/swlh/can-your-business-use-machine-learning-without-data-340c59bf9fb0

Good day!

In my opinion, you need to use all of the methods you listed, and add the selection of features in importance.

If I were you, I would start by choosing an algorithm (Random Forest or Gradient boosting)

Then Feature Engineering

After applied to a method of selecting features according to their predictive power. (significance of features from Random Forest or Deep Feature Selection)

And at the end of Hyper parameter Tuning

However, if the target accuracy is not reached, you can repeat this algorithm iteratively. :)

Mark Panenko Thanks

Naveen Kumar Meena You can use various feature engineering, feature extraction methods, you can find correlation between features and apply machine learning algorithms on that. Further, you can also assign more data to train set and less data to test set or can use k fold cross validation while training the machine learning algorithm.

Saeed Ullah Thnks

You have mentioned you want to improve accuracy without increasing the size of data.

If the reason for not increasing the size of data is unavailability of more data, you may use Data Augmentation (either in the form of transformations such as rotation and translation or augmentations via generative models such as GANs).

The answers to this question may be useful for this topic.