To my understanding, you trained an ANN model and extracted the weights and biases, and now trying to understand what the values mean.
So here are answers to your questions:
Why are some of the weights negative?
- Positive weight implies that the corresponding feature is directly related to the output.
- Negative weight implies that the corresponding feature is inversely related to the output.
Does a larger weight value mean it affects the output parameter more significantly?
- A larger weight implies that the corresponding feature is more important in predicting the output.
Are these weight values random and/or arbitrary?
- No, these weight values are neither random nor arbitrary. At the start of training the neural network, you can initialize these weights either random or arbitrary, or initialize them based on other techniques. But the final weight values which we get after completion of training are neither random nor arbitrary. Final weight values come after the training process is completed and depend on how well the model is trained.