Molecular Machine Learning Kaggle Competition

This repository contains a Jupyter notebook designed for participation in the Kaggle competition: Molecular Machine Learning.

About the Competition

The goal of this competition is to predict the T80 lifetime of organic molecules using a variety of molecular descriptors. The dataset includes both training and test sets, with the test set lacking the target T80 values (as is standard in Kaggle competitions).

Notebook Overview

This notebook walks through a typical machine learning workflow for tabular data, including:

• Data Loading and Exploration:
  • Loads the training and test datasets.
  • Explores data types, missing values, and feature distributions.
• Feature Engineering and Preprocessing:
  • Handles categorical and numerical features.
  • Removes object-type columns and applies scaling and transformation to numerical features.
  • All preprocessing steps (variance threshold, power transform, scaling) are fit on the training set and then applied to the test set using the same fitted transformers, ensuring the test data is in the same format as the training data for valid predictions.
• Model Training and Validation:
  • Trains several regression models (SVM, Lasso, Ridge, Decision Tree, Gradient Boosting, Random Forest, etc.).
  • Performs hyperparameter tuning using grid search and cross-validation.
  • Evaluates models using metrics such as RMSE and MSLE.
  • Visualizes model performance and feature importances.
• Prediction:
  • Generates predictions for the test set (note: true T80 values are not available for the test set).
• Molecule Visualization:
  • Some molecules from the training set are displayed graphically using RDKit.
  • RDKit is an open-source cheminformatics library that allows for the manipulation, analysis, and visualization of chemical structures directly in Python. In this notebook, it is used to convert SMILES strings to molecular structures and visualize them, helping users understand the structural diversity in the dataset.

Citation

David M. Friday, Changhyun Hwang, Seungjoo Yi, Jason L. Wu, Tiara C. Torres-Flores, Martin D. Burke, Charles M. Schroeder, Ying Diao, Nicholas E. Jackson. Molecular Data Machine Learning. https://kaggle.com/competitions/molecular-machine-learning, 2025. Kaggle.

---

For more details, see the notebook file Molecular model.ipynb in this repository.