In machine learning, it is considered that having a large amount of data means having a good-quality model for prediction, but a large dataset also poses the challenge of higher dimensionality (or the curse of dimensionality). It causes an increase in complexity for prediction models due to the large number of attributes. PCA is the most commonly used dimensionality reduction method and helps us to identify patterns and correlations in the original dataset to transform it into a lower-dimension dataset with no loss of information.
The main concept of PCA is the discovery of unseen relationships and correlations among attributes in the original dataset. Highly correlated attributes are so similar as to be redundant. Therefore, PCA removes such redundant attributes. For example, if we have 200 attributes or columns in our data, it becomes very difficult for us to proceed, what with such a huge number of attributes. In such cases, we need to reduce that number to 10 or 20 variables...
