Adapt the neural network for multi-class quality prediction

Transition the target from quality_binary to the full multi-class quality score. Update the output layer nodes, change the loss function to CrossEntropy, and tune the architecture (e.g., adding a second hidden layer or dropout) to maximize the multi-class F1-score.

Start by locating the notebook code that defines the quality_binary target, output layer, and loss function. Evaluate the updated model with the full multi-class quality target and multi-class F1-score while comparing architecture changes such as a second hidden layer or dropout.

Written by the indexing model from the issue text.