Because R-squared statistics are not perfect. In fact, it has one major defect. No matter how many variables we add to the regression model, its value will never decrease.

R-square and adjusted R-square are two evaluation indexes, which are very important for evaluating regression problems. Adding random independent variables does not help to explain the change of target variables. Our r-squared value remains the same. Therefore, giving us a wrong indication, this variable may help to predict the output. However, the decline of the adjusted R-squared value shows that this new variable actually fails to capture the trend of the target variable.