Propose an alternative activation function that can handle larger dynamic ranges and explain its tradeoffs.

Use a scaled tanh activation function, which has a larger range than ReLU and is less prone to saturation. This will improve the model’s ability to detect subtle patterns at the cost of increased computational complexity.