Randomness of training examples in Perceptron leads to decrease in performance

Question

I trained online and the average perceptron model. What I have observed is that my performance decreases if I shuffle my dataset (in between the iterations or even before th