Bias talks about how the model fits the training data. If it fits very bad then its a high bias and vice versa. On the other hand variance mostly checks how well the model performs with the unkown or test data. If accuracy is very much down as compared to the training accuracy then the model is said to have high variance. Ideally we should keep a balance in how much bias we can allow while training the model i.e. even though the model doesn't exactly fit the training data but it can generalize well i.e. for unknown data also it will show similar performance or accuracy i.e. low variance (i.e.accuracy varies less) This balance is called the bias variance trade off.