Hello dear researchers
I need an optimized ensemble-based predictor for a binary problem. Do you suggest a combination of one machine and one deep learning algorithm or a combination of two deep learning algorithms? why? Based on what parameters should be decided?
Thank you for your support
Typically, more than one algorithm is used when employing ensemble learning. These are generally machine learning-based, and not deep learning, but diving into deep learning ensembles is a good idea. As Haseeb Javed
pointed out, combining different base models do indeed produce one optimal predictive model (from my experience working with ensembles anyway). Have a look at the following links (they best answer the question that you posed):1. https://towardsdatascience.com/ensemble-methods-in-machine-learning-what-are-they-and-why-use-them-68ec3f9fef5f
2. https://machinelearningmastery.com/tour-of-ensemble-learning-algorithms/
3. https://machinelearningmastery.com/stacking-ensemble-machine-learning-with-python/ (with code)
4. https://www.analyticsvidhya.com/blog/2018/06/comprehensive-guide-for-ensemble-models/ (with code)
It would be a good idea to introduce boosting algorithms for binary classification tasks, unless it's an image recognition task (although it is possible, ANNs, LSTMs and alike are much better).
You wouldn't generally choose a model based on its parameters (at least I don't). What I generally look for is scalability and effectiveness to model the data. For example, if the dataset chosen was the Iris dataset in excel format, I would typically go for logistic regression or RandomForest, not an ANN. But, if the dataset was significantly large, I'd personally prefer a gradient boosting algorithm or an ANN. The choice is totally up to you, your desired results and what you are going to do with the results after.
Note: tuning the algorithm hyperparameters (be it ensembles or single ML algorithms) sometimes improves the results even more. Hence, it is recommended to tune the hyperparameters for best results.
Good luck with your project!
DEAR Haseeb Javed
Thanks for your guidance as well as your effectiveness.
Good luck
DEAR Shaheen Solwa,
Your answer is very complete and useful
You shared the appropriate links
Thank you very much for your guidance and I will try to choose my model accordingly
Dear Shima Shafiee :
Ensemble methods create multiple models and then combine them to produce improved results. Ensemble methods usually deliver more accurate solutions than a single model would. This has been the case in several machine learning competitions, where the winning solutions used ensemble methods. Which technique can be implemented in this case depends upon your project. Please have a look at these links for further study:
- https://towardsdatascience.com/ensemble-methods-in-machine-learning-what-are-they-and-why-use-them-68ec3f9fef5f
- https://www.analyticsvidhya.com/blog/2018/06/comprehensive-guide-for-ensemble-models/
- https://www.toptal.com/machine-learning/ensemble-methods-machine-learning#:~:text=Ensemble%20methods%20are%20techniques%20that,winning%20solutions%20used%20ensemble%20methods.
regards;
Ehtisham
Shima Shafiee It depends on the complexity of the problem you are working on, so you should start by reviewing the literature on applying different AI and machine learning algorithms to the relevant dataset. As for me, I will always recommend DL algorithms for their promising results for dealing with complex patterns in data.
DEAR Ehtisham Lodhi
Your answer was very promising for my work. Thank you
Voting and averaging are two of the easiest ensemble methods. They are both easy to understand and implement. Voting is used for classification and averaging is used for regression. In both methods, the first step is to create multiple classification/regression models using some training dataset.
https://machinelearningmastery.com/ensemble-methods-for-deep-learning-neural-networks/
- Badges
- Science topic