When explaining PCA, I often find that people struggle to grasp the concept of principal components. They often ask, "What are principal components, again?" This difficulty arises because, in traditional methods, we typically define variables, compute their values, and then interpret the results. In contrast, PCA works the other way around. Rather than starting with predefined variables, PCA extracts linear combinations of the original variables to account the maximum variance in a dataset —these are the principal components. These components can then be interpreted to find meaningful ways, such as identifying bending or torsion modes from the vibration patterns of a tall building.