$ CAT CASE-STUDY.MD · NEURAL NETS VS. BOOSTING
MLP vs. gradient boosting
on regression and classification.
Different multi-layer perceptron architectures, compared head to head with gradient boosting models on regression and classification tasks.
01 · BREAKDOWN
Project breakdown.
Data preprocessing, building and evaluating MLP models of different depths and structures, then comparing them with a gradient boosting regressor and classifier.
Evaluate MLP architectures
Test neural network structures for both regression and classification.
Compare with gradient boosting
Put the best network up against gradient boosting on the same tasks.
Measure what matters
MSE and R² for regression; accuracy and MSE for classification.
02 · FINDINGS
Key findings.
Boosting won on tabular data.
Gradient boosting performed better on both regression and classification. MLPs can be effective, but structured tabular data tends to favor boosting.
Boosting outperformed
The gradient boosting regressor beat the best MLP.
Close, boosting ahead
Both did well; the gradient boosting classifier was slightly better.
Boosting adapts faster
Deep networks were harder to tune; boosting got there with less effort.
