Feature Engineering Case Study in Python

Transform messy raw data into meaningful features to improve a model's generalization, using an iterative cycle of feature creation, model fitting, evaluation, and refinement.

Unlock feature engineering as the unsung hero that prevents garbage in, garbage out by revealing signal in data and driving model performance.

Explore common-sense feature engineering to improve models: clean data, impute missing values, remove outliers, scale features, and create informative new features, including text representations like word vectors.

Explore a feature engineering case study with the Titanic dataset in Python, loading data with pandas and inspecting shape, dtypes, and missing values. Analyze the survival distribution, discuss class balance, and consider downsampling the majority class before modeling.

Explore continuous features with exploratory data analysis, drop columns like passenger id, name, ticket, sex, and cabin, and use describe and correlation to reveal fare and class related to survival.

Explore how continuous features like fare and age relate to survival by grouping by class and survived status, using describe and t-test to compare distributions.

Plot continuous features with seaborn and matplotlib to compare distributions against survivorship, using overlayed histograms for age and fare, plus cat plots to explore class and family effects.

Explore categorical features in the Titanic dataset by assessing missing values, grouping by sex and embarked to compare survival, and engineer a title from names.

Plot categorical features to compare survival rates by title, gender, cabin, and boarding point on the Titanic, using pivot tables and categorical plots.

Finalize 10 features by prioritizing title over name, keep class, sex, and age, combine family features, convert cabin missing into a binary predictor tied to survival, and note embarkation port.

Explore mean value imputation for missing data in a Python feature engineering workflow. Learn filling age with the mean, handling embarked, and creating clean features for a Titanic dataset.

Cap one-dimensional features to remove outliers using the 95th and 99th percentiles and three standard deviations, applying z-scores to age, fare, and other features.

Transform skewed features in Python with Box Cat power transformations to tighten tails. Evaluate with cucu plots and normal-curve histograms, then apply the exponent one over five to fair feature.

Derive a new title feature from text by parsing the name field, transform data for modeling, and explore how the title correlates with survival to improve feature engineering in Python.

Create a binary indicator for missing cabin data in the Titanic dataset, revealing its correlation with survival and clarifying model signals in a Python feature engineering case study.

Combine siblings and spouses aboard with parents and children aboard to create a single family count feature, store it, and test its impact in the modeling phase.

Convert categorical features to numeric using laybourne coding, mapping male to one and female to zero, fit and transform data, and prepare features for modeling.

Split a full dataset into training, validation, and test sets using two passes of train_test_split to achieve 60/20/20, preparing features and labels for unbiased model evaluation on unseen data.

Learn how to scale features using a standard scaler to convert data to z-scores, align features on a common scale, and transform training, validation, and test sets for model performance.

Create datasets for modeling across four feature sets—original, clean, engineered, and reduced—using pandas to train, validate, and compare models for survival prediction.

Fit a model on raw original features, use grid search with cross-validation to select parameters, and assess feature importances with a random forest in Python, aided by correlation heat maps.

Compare cleaned features with original features by handling missing values and outliers, using the correlation matrix insights, to reveal performance benefits, including 84.7% accuracy and similar feature importance.

Explore how clean and transformed features, including the cabin indicator and family count, enhance model performance through correlation analysis and grid search, achieving 83.7% accuracy.

Reduce feature sets, examine correlations and feature importance, and build a model. The eight-feature model with depth eight achieves eighty three point one percent accuracy, driven by sex and title.

Evaluate four feature-set models in this python feature engineering study on the validation set, measuring accuracy, precision, recall, and latency, then confirm the best model on the test set.