Titanic Survival Neural Network

Problem

The Titanic dataset is a classic benchmark for practicing feature engineering and neural network design on tabular data with mixed types.

Approach

Built a neural network with a tf.data pipeline and TensorFlow feature columns to predict passenger survival from the Titanic dataset.

Tech highlights

TensorFlow 2 / Keras with a tf.data input pipeline, categorical encoding via feature columns, and engineered features including age groups, fare brackets, and a family-size composite.

Results

Trained a working survival-prediction model with engineered features and categorical encoding.

Challenges

Preventing the network from overfitting such a small, noisy dataset, handling missing age and cabin values, and correctly wiring mixed categorical and numeric features through the tf.data pipeline without leaking information between the train and validation splits.