Marginal likelihood
it is computed by marginalizing over (or integrating out) the unknown variable . This can be interpreted as the average (with respect to the prior) probability of the data. And this is a constant term.
In other words, evidence acts as a normalization constant in Bayes' rule. It plays the role of filtering from the whole in the total probability.
Evidence should not determine belief, but mathematically update it. That is to say, when assessing a situation, considering the probability prior to the evidence along with the evidence itself leads to greater accuracy.
