Comparisons
Pairs that get conflated in real metric reviews and real design docs — batch and online, precision and recall, data drift and concept drift, validation and test. Neither column wins; what decides is the problem. Each record leads with the confusion, because the confusion is the reason the record exists.
Data drift vs concept drift
Both get called "drift" and both trigger the same reflex — retrain — when they need different responses and are detected by different signals. Data drift is visible without labels: compare feature distributions week to week. But it is not failure; a model can be perfectly correct on a shifted population if the relationship it learnt still holds, and the drift explorer in this domain has scenarios where features move and quality does not. Concept drift is invisible without labels: the features are stable, the predictions are stable, and only when outcomes arrive does the error appear — which makes ground-truth delay the constraint on detecting it. The strongest form of "just retrain on drift" is that on a fast-moving problem with cheap training and fast labels, it is a fine default. The strongest form of the opposite is that retraining on data-drifted inputs with a feature bug bakes the bug in, and retraining on concept drift only works if the new labels reflect the new concept. The honest sequence is: which signal fired, is quality actually down on the drifted slice, and what changed in the world or in the pipeline — then decide.
As the diagnosis when feature monitors fire and the model is being asked about regions of input space it rarely saw: a new country, a new device, a changed form, a marketing campaign that shifted who arrives.
As the diagnosis when the inputs look the same but quality fell on labelled outcomes: fraudsters adapted, a competitor changed prices, a policy changed what the label means.
| Dimension | Data drift — the distribution of the inputs moved | Concept drift — the relationship between inputs and target moved |
|---|---|---|
| What moved | P(X) | P(Y | X) |
| Detectable without labels | Yes | No |
| Typical cause | New population, new channel, upstream change | Adaptation, policy change, market change |
| Does quality necessarily fall | No — often harmless | Yes, by definition |
| First response | Check for a feature bug; measure quality on the slice | Confirm with labels; look for what changed |
| Does retraining fix it | Only if the new region has labels and no bug | Only if new labels reflect the new relationship |
| Monitor that catches it | Feature distributions | Outcome metrics, delayed |
Model families compared
Linear, trees, boosting, neural and k-NN — compared without naming a winner, and with the block that says where the comparison stops being true.
No column is a winner. Each family is a set of assumptions about the data — linear separability, axis-aligned interactions, additive residuals, a representation that can be learned, locality in feature space — and the right one is the one whose assumptions your data happens to satisfy at the size you have. The reflex answer “XGBoost” is the red flag this table exists to catch: it is often right on tabular data and it is never right as a reflex, because it skips the baseline that would have told you whether anything more than a linear model was needed. The where this comparison misleads block on every row is the part worth reading.
Count independent entities, not rows — a million events from ten thousand users is ten-thousand-sized data for generalising to new users. And the neural column flips completely with transfer learning: a pretrained model fine-tuned on two thousand images beats every other column on that task.
The "boosting wins on tabular" claim is true for a tuned model on tens of thousands of clean rows with a validation set. It is not true for three hundred rows, for a problem whose signal is linear, for a regulator who wants coefficients, or for a team with no time to tune — and the gap to the linear model is often inside the error bar.
The columns are not competing on the same input. Once a pretrained network has produced an embedding, a linear model or k-NN on top of it is often within a few points of full fine-tuning — so "neural for images" usually means "a neural representation, then whichever head is cheapest".
Interpretability is not one property. A linear model with two hundred correlated features and L1 selection is harder to explain honestly than a depth-three tree; and every column's "explanation" is undermined equally by a leaked or proxy feature, which the explanation will present with confidence.
Training cost is dominated by the number of runs, not the run: a boosted model tuned over two hundred configurations costs more than a network fine-tuned once. And the k-NN column's zero is a loan repaid on every query.
The model is rarely the slow part. Feature retrieval, a network hop and JSON serialisation usually dwarf any of these numbers, so a latency budget is a question about the serving path before it is a question about the family.
A low burden on the modelling side does not remove the burden on the serving side: every column's features must be reproduced at prediction time with the same code, the same freshness and the same point-in-time semantics. Trees remove the need to engineer features, not the need to serve them.
Native handling is convenient and dangerous in equal measure: it lets a tree model learn that "missing" predicts the target, which is fine until serving produces missingness for a different reason — a timeout, a new form — and the model reads the outage as a signal.
Calibration only matters if a downstream decision multiplies the score by a cost or compares it to a probability threshold; a pure ranker does not need it. And calibration measured on the validation set drifts with the base rate in production, for every column alike.
Neither behaviour is "correct". A tree that predicts last year's maximum for a record-breaking day is wrong; a linear model that predicts a negative price is wrong differently. The honest answer is a monitor on inputs outside the training range and a fallback for them, whichever family serves.
Irrelevant is not the danger — leaky is. Every column will seize a feature that carries the answer, and the more capable the model the more efficiently it does so; robustness to noise says nothing about robustness to leakage, which only a point-in-time audit provides.
Every column is wrong somewhere, and the question that finds where is the same for all of them: how much data, of what modality, at what latency, with what interpretability requirement, judged by which metric, at what cost to train and serve. A model family chosen before those are answered is a guess with a library name.