Leon MutisyaFeature Engineering
Feature Engineering is described as a preprocessing step in machine learning which transforms raw data into a more effective set of inputs which have several attributes known as features.
The success of machine learning models heavily depends on the quality of the features used to train them. Feature engineering involves a set of techniques that enable us to create new features by combining or transforming the existing ones. These techniques help highlight the most important patterns and relationships in the data, which in turn helps the machine learning model to learn from the data more effectively.
Key Techniques in Feature Engineering
Feature Engineering can be classified into two key steps namely;
Data Pre-processing
Business Understanding(Domain Knowledge)
Data Pre-processing
This is usually a step in feature engineering and involves preparing and manipulating the data to the current machine language needs. Various techniques are used here among them;
Handling Missing Value where techniques like imputation (mean, median, mode), or using algorithms that handle missing values natively can be employed.
Encoding Categorical Variables where categorical data must be converted into numerical form for most algorithms using common methods like one-hot encoding, label encoding, and target encoding.
Scaling and Normalization where scaling features ensures that they contribute equally to the model. Techniques include standardization (z-score)
Feature Interaction & Feature Creation where existing features are combined to create new features thus creating complex relationships with the data
Dimensionality Reduction where techniques like PCA (Principal Component Analysis) or t-SNE reduce the number of features while retaining the most important information.
EDA can also be utilized in feature engineering and is usually a precursor to feature engineering.
Domain Knowledge
Domain knowledge refers to the understanding and expertise in a specific field or industry. In feature engineering, it involves applying insights and understanding of the data's context and relationships to create meaningful features that can enhance model performance.
It helps in identifying which features are relevant to the problem at hand and understand data relationships.
