Feature Scaling Comparison Lab for Standard, MinMax, and Robust Methods

Distance-based and gradient-based models quietly assume features live on comparable scales, and violating that assumption is one of the most common silent errors in beginner machine learning. Choosing between standardization, min-max normalization, and robust scaling requires understanding what each does to a distribution — especially when outliers are present.

The intern builds a Streamlit playground where users upload a dataset and apply scikit-learn's StandardScaler, MinMaxScaler, and RobustScaler to selected features, with before-and-after distribution plots rendered side by side. The app makes the differences tangible: how standardization recenters data around zero, how min-max compresses everything into a fixed range, and how robust scaling shrugs off outliers that badly distort the other two. Users can inject artificial outliers and watch each technique respond differently, turning an abstract preprocessing choice into an observable behavior.

The project turns preprocessing from a memorized ritual into an understood decision. The intern learns when each scaling method is appropriate, how scaling interacts with downstream models, and how to build a compact interactive Python tool that demonstrates a statistical concept clearly enough to teach it.