Dependence numbers close to one indicate that the feature is completely predictable using the other features, which means it could be dropped without affecting accuracy. For example, the mean radius is extremely important in predicting mean perimeter and mean area, so we can probably drop those two. It also looks like radius error is important to predicting perimeter error and area error, so we can drop those last two. Mean and worst texture also appear to be dependent, so we can drop one of those too. Similarly, let's drop concavity error and fractal dimension error because compactness error seems to predict them well. Worst radius also predicts worst perimeter and worst area well.