Binning, Log, and Power Transforms — Data Analysis & Feature Engineering

Transforming the distribution of a numeric feature — through binning, logarithms, or power transforms — can reshape skewed, heavy-tailed, or non-linear variables into forms that models handle far better. These transforms exist because many algorithms perform best when features are roughly symmetric or when a non-linear relationship is linearised, yet real data is full of right-skewed quantities like income, prices, and crowd sizes that violate these preferences. Binning discretises a continuous variable into categories, the log transform compresses a right-skewed variable's long tail, and power transforms like Box-Cox and Yeo-Johnson systematically find the transformation that best normalises a distribution. Each reshapes the feature differently and suits different situations, and knowing when to discretise, when to compress, and when to systematically normalise is what lets an analyst tame the awkward distributions that otherwise degrade model performance.

Analogy🏏Cricket

🏏 Think of it like cricket: Reducing twenty batting statistics to two dimensions with PCA produces a flat projection that may look like one undifferentiated cloud, while t-SNE or UMAP finds that the data actually organises into distinct clusters — aggressive pinch-hitters, steady anchors, explosive finishers — whose separation PCA's flat projection smeared together. Just as the non-linear reduction reveals the natural groupings that flat projection could not, t-SNE and UMAP reveal structure that PCA misses because it can only flatten, not curve to follow the data's natural shape. The insight is that non-linear reduction methods follow the data's true curved geometry rather than forcing a flat projection, revealing the cluster structure and local neighbourhoods that linear methods like PCA cannot preserve.