data science

accuracy

total correct/total predictions

if 95% passengers died, predicting died every time gets 95% accuracy but catches zero survivors

TP+TN / TP + TN+ FP + FN

precision

of everything you predicted as positive/flagged how many actually were positive. use when false positives are costly, like flagging a trade as fraudulent

TP/ TP + FP

recall (sensitivity)

of all actual postivits, how many did you catch. high recall = fewer misses.

use when false negatives are costly, like missing a client or cancer

F1 Score

harmonic mean of precision and recall, use when you need to balance both.

AUC - ROC

measures hwo well the model seperates classes across possible thresholds. 0.5 = random guessing, and 1 = perfect

confusion matrix

the 2×2 table behind everything

RMSE (root mean squared error)

average prediction error in teh same units as the target

penalizes large errors heavily bc of the squaring

big misses are much worse than small ones

MAE mean absolute error

treats all error equally, more robust to outliers than rmse

if rmse is much higher than mae, then you have large outliers in predictions

R²

proportion of vairance explained by the x variables.

1 = perfect, 0 = not good

roc curve

AUC = Area under the curve, 0.5 means the model is gussing randomly and 1 emans perfect

Interpretation of AUC:

• 1.0: Perfect Classifier.

• 0.9–0.99: Excellent discrimination.

• 0.8–0.89: Good discrimination.

• 0.7–0.79: Fair discrimination.

• < 0.7: Poor/Weak discrimination.

• 0.5: No better than random chance.

Bias-variance tradeoff

high bias (underfits), complex model = high variance (overfits)

Regularization

L1 (Lasso) drives coefficients to zero (feature selection), L2 (Ridge) shrinks them (prevents overfitting)