Recommended Free Tools
On one synthetic admissions dataset, five classifiers scored between 90.50% and 93.00% accuracy on the same 200-row test set. That narrow spread does not establish a universal best model. The more revealing result was the Decision Tree: it scored 100.00% on training data but 90.50% on test data, a clear warning that training accuracy alone can mislead.
What was compared?
Ayush S Pangaonkar’s DEV Community article, “Five Classifiers, One Dataset: What I Learned About Model Choice”, describes a binary prediction task: estimate whether a student is admitted using an entrance exam score and a percentage. The experiment used 1,000 synthetic students and two features drawn uniformly from 0 to 100.
As an Amazon Associate I earn from qualifying purchases.
For each row, the author calculated an admission probability as 1 / (1 + exp(-(0.1*X1 + 0.2*X2 - 10))), then sampled the label from a Bernoulli distribution using that probability. The resulting labels comprised 755 admitted students and 245 not admitted.
The author split the data into 800 training rows and 200 test rows, using random_state=42. All five models used the same split, making their reported scores comparable within this particular experiment.
#1 Best Overall
How did the five classifiers score?
The figures below are those reported by Pangaonkar for this generated dataset and split. They are not independently reproduced results or general benchmarks.
| Classifier | Training accuracy | Test accuracy | F1 for admitted students |
|---|---|---|---|
| Logistic Regression | 94.13% | 92.50% | 95.05% |
| SVM (RBF kernel) | 93.75% | 91.50% | 94.46% |
| Decision Tree | 100.00% | 90.50% | 93.77% |
| Random Forest (100 trees, entropy) | 100.00% | 93.00% | 95.45% |
| KNN | 95.50% | 92.00% | 94.81% |
What does the training-to-test gap show?
The Decision Tree’s perfect training score fell to 90.50% on the test set. That pattern is consistent with overfitting: a model fits the examples it learned from exceptionally well but performs less well on unseen examples. Pangaonkar describes it as memorizing the training set. Random Forest also reached 100.00% training accuracy, but its test accuracy was higher at 93.00%; a perfect training score by itself does not determine how well a model generalizes.
Logistic Regression had a smaller difference between training accuracy (94.13%) and test accuracy (92.50%). The gap is useful to inspect alongside test performance, but one split cannot establish how stable that performance would be on other samples.
Is there a clear winner?
No. Test accuracy ranged from 90.50% for the Decision Tree to 93.00% for Random Forest, a 2.50-percentage-point range. The test set had 200 rows, so a single changed prediction corresponds to 0.50 percentage points. Small score differences deserve caution at this sample size, especially because the experiment reports just one split and no repeated-run results or uncertainty intervals.
Random Forest had the highest reported test accuracy and F1 for admitted students in this table, but that does not make it a proven winner. Pangaonkar’s own conclusion is appropriately restrained: “I would not call Random Forest a clear winner over Logistic Regression from this experiment alone.”
Why the synthetic data matters
The labels were generated from a logistic-function probability. That construction gives Logistic Regression a favorable match to the task’s underlying form, so its strong result is plausible in this setup. It does not show that real admissions decisions follow the same formula, nor does it establish how these models would perform on real student data.
Rank #4
When comparing classifiers for a real task, ask whether the data and label-generation process resemble the problem you actually need to solve. A model that performs well on this synthetic setup may not remain the best choice when the features, relationships, class balance, or data quality change.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
What should you examine beyond accuracy?
The source reports F1 for the admitted class, which combines precision and recall into one measure. It does not give a full confusion matrix or separate precision and recall values, so those cannot be inferred from the table alone.
The class counts are uneven: 755 students were admitted and 245 were not. For an application where missing an admission or incorrectly predicting one has different consequences, examine precision and recall for each class as well as accuracy. The relevant trade-off depends on what errors matter in the intended use.
How to make a stronger comparison
This experiment is a useful illustration of comparing models on common data, but a more dependable model-choice decision needs evidence about variability and the task itself.
- Keep evaluation data separate. Compare candidates on data they did not use to fit their parameters, and keep the same evaluation procedure across candidates.
- Repeat evaluation across splits or folds. Report how scores vary, not only one result. This helps reveal whether a small apparent advantage persists across samples.
- Choose metrics for the cost of errors. Report accuracy alongside class-specific precision and recall, and use F1 where a combined precision-recall summary is useful.
- Check that the evaluation reflects the real task. Synthetic data can help demonstrate a method, but decisions about deployment require representative data and an evaluation design suited to the real setting.
The article’s experiment illustrates two durable lessons: compare models under the same conditions, and judge generalization using held-out performance rather than training scores alone. Its results are a report about one synthetic dataset—not a ranking of classifier families.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsQuick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
