Developed a neural network dropout customization API for the sktime/sktime repository, focusing on enhancing regularization control in deep learning models for time-series tasks. The solution introduced a flexible dropout parameter for both MLPNetwork and RNNNetwork, allowing users to specify either a uniform float or a per-layer tuple, while maintaining backward compatibility with previous hardcoded defaults. Implemented in Python using Keras, the changes included robust validation to ensure tuple lengths matched network layers and direct passthrough to Keras SimpleRNN. This work addressed API stability, resolved issue #9103, and streamlined experimentation and onboarding for machine learning practitioners using sktime.