There is no fixed rule for determining the exact number of hidden layers, but you can consider the following guidelines and practices:

The important thing is the accuracy of the model. And the accuracy of the model can be effected by changing the number of layers. But in case of small dataset, the model accuracy also changes by set.seed where you shuffle the data for testing and training. However in case of reasonable dataset, the effect of set.seed is negligible.

Selecting the appropriate number of hidden layers in a deep learning model is more of an art than an exact science. It primarily depends on the complexity of the problem and the amount of data you have. There are no strict rules, but here are some general guidelines:

1.For simpler problems, fewer layers are typically better: Simple problems often don't require deep networks, and you may find that one or two layers are enough.

2. For more complex problems, more layers may be necessary: For complex tasks such as image recognition, natural language processing, or any task with a high degree of complexity, more hidden layers are generally beneficial. State-of-the-art models for these tasks often use many layers (sometimes hundreds), but remember that these models also require a lot of data and computational power to train effectively.

3. Overfitting vs Underfitting: If you have a smaller amount of data, using too many layers can lead to overfitting, where the model learns the training data too well and performs poorly on new, unseen data. Conversely, using too few layers can result in underfitting, where the model is too simple to capture the underlying patterns in the data.

4. Practical Considerations: More layers mean more parameters, which means you'll need more computational power and memory. You also need to consider the time it takes to train the model. Deeper networks typically require more time to train.

5. Use Pretrained Models: For many tasks, it's common to use pre-trained models that have already been trained on a large dataset. These models often have a large number of layers, but since they're pre-trained, you can leverage their existing knowledge and just fine-tune them on your specific task.

6. Experiment: Deep learning is a field that benefits greatly from experimentation. Don't be afraid to try different architectures and see what works best for your specific problem.

Lastly, there's ongoing research in developing methods that can automatically determine the optimal network architecture for a given task, known as AutoML or neural architecture search (NAS). These techniques can automatically determine the number of layers, among other hyperparameters, but they require a large amount of computational power and are still an active area of research.

There is no rule for fixing the number of hidden layers. But one can consider the following things while choosing the number of hidden layers.

(i) The complexity of the program

(ii) The size of the dataset

(iii) Available resources

There isn't a set formula for calculating the ideal number of hidden layers. The choice of the number of hidden layers may significantly impact the performance of a deep learning model. Although the following principles and factors might help to make a well-informed decision:

1. Problem Complexity

2. Data Availability

3. Model Capacity

4. Computational Resources

5. Transfer Learning

6. Regularization Techniques

7. Empirical Evaluation

The ideal design might change based on the precise dataset and problem. It's good to start with a basic architecture and progressively add hidden layers to make it more sophisticated while keeping track of the model's performance.