Feature Engineering using R

To summarize, it's important to spend time extracting, selecting and constructing features based on your data and it's size. It will be valuable in improving the performance of your model. Some features could be selected from a given feature set based on correlation with other predictors or with the label. If you have a small set of features, a quick brute-force approach of attempting different combinations might give you the best set of predictors. If you have a huge feature set especially when compared to the total number of data points – exploring dimensionality reduction techniques or other methods to combine multiple features together might be of help. But ultimately, though some of the methods above would help pick features, it's up to you to evaluate their value in being good predictors.