Feature Engineering

The algorithms we use in machine learning will perform based on the quality and goodness of the data; they do not have any intelligence of their own. The better and innovative you become in designing features, the better the model performance. Feature engineering in many ways helps in bringing the best out of data. The term feature engineering essentially refers to the process of the derivation and transformation of given features, thus better characterizing the meaning of the features and representing the underlying problem of the predictive model. By this process, we anticipate the improvement in the model's predictability power and accuracy.

Discretization

In Chapter 3, Introduction to Supervised Learning, we converted the numeric values of a 3-hour rolling average of PM2.5 in the Beijing dataset to the binary values 1 and 0 for logistic regression, based on the threshold of 35, where 1 means normal and 0 means above normal. The process is called discretization, also...