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.
ROC AUC vs PR AUC
ROC AUC is read as a general-purpose quality score, and on an imbalanced problem it flatters. The false-positive rate divides by the negatives; with a million negatives, ten thousand false alarms is a rate of one percent, so the ROC curve hugs the top-left while nine out of ten flags are wrong. PR AUC divides by the predicted positives instead and shows that immediately. The strongest form of the ROC side is that it is base-rate-independent — the same model scores the same ROC AUC whether the positive rate is one percent or thirty, which makes it the right number for comparing models across datasets and for saying how well the scores rank in general. The strongest form of the PR side is that it measures what an operator experiences: the fraction of flags worth acting on at each recall. Neither is a decision; both are summaries over every threshold, most of which you will never use. The number that maps to the decision is precision and recall at the threshold you will actually deploy, and either AUC is a way to compare candidates before that threshold is chosen.
When the classes are reasonably balanced, when you care about ranking across the whole population, or when comparing models on the same data where the base rate is not the point.
When positives are rare and the question is how well the model finds them — fraud, defects, rare diagnoses — because the false-positive rate stays tiny even when most flags are wrong.
| Dimension | ROC AUC — true-positive rate against false-positive rate, threshold-free | PR AUC — precision against recall, threshold-free, positive class only |
|---|---|---|
| Axes | TPR against FPR | Precision against recall |
| Sensitive to base rate | No — a feature and a flaw | Yes — reflects the actual positive rate |
| Under heavy imbalance | Optimistic; FPR stays small | Honest about wasted flags |
| Random classifier scores | 0.5, always | The positive rate |
| Interpretation | P(random positive ranked above random negative) | Average precision over recall levels |
| Best for | Comparing rankers across datasets | Rare-event detection with a review cost |
| Neither gives you | A threshold | A threshold |
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.