Making Predictions with All Three Models

End-to-End Machine Learning: Titanic Survival Prediction

2 min read

Published Nov 18 2025, updated Aug 17 2026

10
0
0
0
KerasMachine LearningMatplotlibNumPyPandasPythonscikit-learnSciPySeabornTensorFlow

Once the models are trained and evaluated, a common final step is to use them for real predictions.


We will prepare an example passenger (or multiple passengers) and run them through:

  • Logistic Regression
  • Random Forest
  • Keras Neural Network

This demonstrates how to make new, real-world predictions using your pipelines and transformed data.






Create a Sample Passenger to Predict

We’ll create a small dataframe with one or more new passengers.

Here’s an example with two fictional passengers:

# Example passengers to predictsample_passengers = pd.DataFrame([    {        "pclass": 1,        "sex": "female",        "age": 25,        "sibsp": 0,        "parch": 0,        "fare": 100,        "embarked": "S",        "class": "First",        "who": "woman",        "alone": True    },    {        "pclass": 3,        "sex": "male",        "age": 35,        "sibsp": 1,        "parch": 3,        "fare": 12,        "embarked": "S",        "class": "Third",        "who": "man",        "alone": False    }])​sample_passengers

In tabular form:

   pclass     sex  age  sibsp  parch  fare embarked  class    who  alone0       1  female   25      0      0   100        S  First  woman   True1       3    male   35      1      3    12        S  Third    man  False

Passenger A:

  • First class young woman travelling alone

Passenger B:

  • Third class adult man with family

We expect dramatically different survival probabilities.






Predictions with Logistic Regression

pred_lr = log_reg.predict(sample_passengers)prob_lr = log_reg.predict_proba(sample_passengers)[:, 1]​print("Logistic Regression Predictions:", pred_lr)print("Logistic Regression Probabilities:", prob_lr)

Output:

Logistic Regression Predictions: [1 0]Logistic Regression Probabilities: [0.94124233 0.0275436 ]

The output gives:

  • 0 = predicted death
  • 1 = predicted survival
  • Probability = model’s confidence the passenger survives





Predictions with Random Forest

pred_rf = rf.predict(sample_passengers)prob_rf = rf.predict_proba(sample_passengers)[:, 1]​print("Random Forest Predictions:", pred_rf)print("Random Forest Probabilities:", prob_rf)

Output:

Random Forest Predictions: [1 0]Random Forest Probabilities: [0.995 0.015]

Random Forest typically gives:

  • Stronger confidence for clear “rules”
  • Slightly noisier for edge cases





Predictions with the Keras Neural Network

The Keras model requires preprocessed numeric arrays, so we must transform using the fitted preprocessor:

sample_processed = preprocessor.transform(sample_passengers)​# Convert sparse matrix if neededif hasattr(sample_processed, "toarray"):    sample_processed = sample_processed.toarray()​prob_keras = model.predict(sample_processed).ravel()pred_keras = (prob_keras >= 0.5).astype(int)​print("Keras Predictions:", pred_keras)print("Keras Probabilities:", prob_keras)

Output:

Keras Predictions: [1 0]Keras Probabilities: [0.94335914 0.07789665]





Final Combined Output

results = pd.DataFrame({    "Passenger": ["A (1st class woman)", "B (3rd class man)"],    "LR_Prob": prob_lr,    "LR_Pred": pred_lr,    "RF_Prob": prob_rf,    "RF_Pred": pred_rf,    "Keras_Prob": prob_keras,    "Keras_Pred": pred_keras})​print(results)

combined prediction results

Passenger A (1st-class woman travelling alone)

  • All three models give a very high survival probability (> 94%).
  • Random Forest is almost certain (99.5%).
  • Logistic Regression and Keras also strongly agree.
  • This aligns with:
    • “Women first” evacuation procedures
    • First-class passengers having better access to lifeboats
    • Historical survival rates (≈97% for 1st-class females)

All models predict survival confidently.


Passenger B (3rd-class man with family)

  • All three models give a very low survival probability (< 8%).
  • Random Forest gives near-zero chance (1.5%).
  • Logistic Regression agrees (≈2.7%).
  • Keras is slightly less certain but still low (7.8%).
  • This matches historical data:
    • Only ~14% of 3rd-class males survived
    • Men were deprioritised
    • Family groups in 3rd class were often delayed or blocked from access

All models predict he does not survive.

© 2025 SimpleSteps.guide
AboutFAQPoliciesContact
End-to-End Machine Learning: Titanic Survival Prediction | Making Predictions with All Three Models | SimpleSteps.guide