1. Be precise about the analysis method and the details used to analyze data.
It's not enough to state that you use machine learning, artificial intelligence, or neural networks. This is simply too vague.
We need specific details about which analysis method will be used, why it has been chosen, to assess whether it is relevant for the particular data.
2. How are the results assessed, and what is considered satisfactory?
Describe in the application how you will measure if there is an effect of an algorithm, and also how significant this effect should be to consider the result satisfactory.
For example, if an algorithm is used to analyze X-ray images, such as deep neural networks, it's not enough to say that this is just the method used. You must also explain what you are looking to find in these images, how you measure it, and how significant the effect should be when using the particular algorithm.
3. Avoid overfitting
In the application, it's important to account for the precautions you plan to take to avoid overfitting. This could be, for example, dividing your dataset into a training set, a validation set, and a test set, or other extensions to the analysis methods that prevent overfitting.
4. Account for sample size
It's still necessary to account for the sample size with these advanced methods, and we can use the same ideas as for traditional analysis methods. For example, how precise should a prediction be, or what level of precision is desired to make a given prediction.
It's necessary to provide an explanation of how large the dataset you need is and how this has been measured, so it can be assessed whether there is a sufficient data basis to carry out the project.
5. Who will conduct the analyses?