1. The Why
After all the cleaning and feature engineering, your dataset is ready — but data alone is useless. The real test: can you learn real patterns and make reliable predictions?
2. The What (Plain English)
Start with a simple baseline: Logistic Regression. Three critical steps:
- Split the Data — hide part as a test set for honest evaluation.
- Train (
.fit()) — show the model the training set to learn patterns. - Predict & Evaluate — test predictions on the hidden set.
3. The How (Lab Notebook)
Step A: Separate Features & Target
Split your DataFrame into input features X and target y.
Step B: Train-Test Split
Use train_test_split (20% test, random_state=42).
Step C: Train & Predict
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
X_train, X_test, y_train, y_test = train_test_split(X_scaled, y, test_size=0.2, random_state=42)
model = LogisticRegression()
model.fit(X_train, y_train)
y_pred = model.predict(X_test)Step D: Evaluate
-
Local test accuracy: 0.88
-
Classification: high recall for non-survivors (0.95), moderate for survivors (0.75).
Step E: Kaggle Submission
Ran the pipeline on test.csv. Public score: 0.775 — a normal “generalization gap.”
4. Academic Bridge
5. Why It Matters
Building a model is easy. Explaining trade-offs and the generalization gap is the mark of a real data scientist. This is what employers care about.