Smoker Binary Prediction using Bio-Signals > Go to project

Overview:

This notebook analyses the ‘Binary Prediction of Smoker Status using Bio-Signals’ dataset from the Kaggle Plagounded Series - Season 3, Episode 24. Link below.

https://www.kaggle.com/competitions/playground-series-s3e24/overview

Aim:

Use binary classification to predict a patient’s smoking status given information about various other health indicators.

Evaluation Metric

Submissions are evaluated on ‘area under the ROC curve’ between the predicted probability and the observed target.

Improve Data Quality

• Convert CSV data into a Pandas DataFrame.
• Remove irrelevant or unimportant data.
• Convert appropriate data into numerical format for improved usability.
• Rename columns and features for better clarity.
• Introduce new columns and features to enhance analysis.

Observations

• Identify outliers in the data.
• Summarize key aspects of the dataset.
• Analyse trends and patterns, including revenue and student enrollments.

Visualistions

• Create visualisations using the Matplotlib and Seaborn libraries.
• Present findings to showcase financial trends and significant insights.

Model Creation

• Created baseline models for early predictions.
• Selected best performing models (XGBoost, Catboost, LightGBM & Random Forest) to improve through CV.
• Used best models to create a Neural Network Stacking Classifier with Scikeras’s KerasClassifier and Scikit-Learn’s StackingClassifier

Conclusion

• Saved the cleaned and transformed data for future analysis.
• Saved the model for future use.
• Create a predictions CSV file for competition submision.
• Our final model’s predictions finished with a score of 0.87583, which placed 283 out of 1910 submissions.