KO | EN

GitLite

Minimalist Open Source Repository Search Engine

gitlite — search

#typescript #ai-agents #developer-tools #claude-code #codex #cli #agent-skills #ai #dsh-plugin #mcp #windows #agent

Kaggle_Silver_Medal_Solutioun_CMI-PIU

★ 55 Open GitHub ↗

This project aims to analyze physical activity data from children and adolescents to predict the extent of their problematic internet use. The goal is to develop a model that can help identify early signs of problematic internet use, thereby promoting the development of healthy digital habits.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

## Child Mind Institute — Problematic Internet Use (CMI‑PIU) Solution
![Python](https://img.shields.io/badge/python-3.10+-blue.svg)

This solution was developed for the Kaggle competition [Child Mind Institute — Problematic Internet Use](https://www.kaggle.com/competitions/child-mind-institute-problematic-internet-use), where participants predict the severity category `sii ∈ {0,1,2,3}` using demographics, clinical/questionnaire data, and wearable time‑series signals. The approach combines tabular features with per‑participant time‑series summary statistics and trains tree‑based regressors with out‑of‑fold validation and threshold optimization for the Quadratic Weighted Kappa (QWK) metric.

Our work earned a Silver Medal. The repository contains an end‑to‑end notebook that reproduces training and inference and generates a valid `submission.csv`. 🥈

For a detailed walkthrough of the solution and insights into the modeling process, see my [blog post](https://chenfenghuang.info/2024/12/20/Kaggle-CMI-PIU/).
![CMI‑PIU — Silver Medal](https://raw.githubusercontent.com/chenfeng-huang/Kaggle_Silver_Medal_Solutioun_CMI-PIU/main/fig/certificate.png)

## Competition Overview

### Competition Introduction

The task is to build a predictive model for `sii` (four ordered categories) using a mixture of tabular data (demographics, physical health, and questionnaires such as SDS/PCIAT) and multivariate time‑series recorded per participant. The evaluation metric is **Quadratic Weighted Kappa (QWK)**, which measures agreement between predicted classes and ground truth while penalizing larger class discrepancies more heavily.

### Competition Background

Problematic Internet Use is an emerging mental‑health concern. The dataset couples questionnaire and clinical information with wearable time‑series, enabling models that can leverage both static and dynamic signals. A practical solution must align columns across train/test, summarize long time‑series efficiently, and handle missing/categorical features robustly while optimizing directly for the ordered‑class QWK objective.

## Solution Overview

This solution comprises three components:

1. **Time‑Series Feature Extraction**: Aggregate each participant’s parquet time‑series into compact summary statistics via `DataFrame.describe()` (e.g., count, mean, std, quantiles, max) per channel.
2. **Tabular Fusion and Preprocessing**: Merge time‑series stats with tabular features by `id`, fill and encode season‑type categoricals, and align train/test columns.
3. **Modeling and Thresholding**: Train tree‑based regressors with 5‑fold stratified CV on `sii`, ensemble predictions where helpful, and optimize three thresholds to discretize continuous predictions into {0,1,2,3} maximizing QWK.

The final output is a two‑column `submission.csv` with `id,sii`.

## How to Reproduce

### Environment Setup

- Python 3.10+ recommended
- Install dependencies:

```bash
pip install -r requirements.txt
```

Key libraries: `numpy`, `pandas`, `scikit-learn`, `lightgbm`, `xgboost`, `catboost`, `scipy`, `pyarrow`, `tqdm`.

### Data Layout

Place the competition data under `dataset/`:

```
dataset/
├─ train.csv
├─ test.csv
├─ sample_submission.csv
├─ series_train.parquet/
│  └─ id=<participant_id>/part-0.parquet
└─ series_test.parquet/
   └─ id=<participant_id>/part-0.parquet
```

### Data Source and Citation

- Dataset and rules: see `dataset/data_source.md` and the Kaggle data page: https://www.kaggle.com/competitions/child-mind-institute-problematic-internet-use/data

```bibtex
@misc{child-mind-institute-problematic-internet-use,
    author = {Adam Santorelli and Arianna Zuanazzi and Michael Leyden and Logan Lawler and Maggie Devkin and Yuki Kotani and Gregory Kiar},
    title = {Child Mind Institute — Problematic Internet Use},
    year = {2024},
    howpublished = {\url{https://kaggle.com/competitions/child-mind-institute-problematic-internet-use}},
    note = {Kaggle}
}
```

### Training and Inference

- Launch Jupyter and open the notebook:

- Run all cells in `cmi-piu-silver-medal-solution.ipynb` to train, validate, and generate `submission.csv` in the repository root.

## Author

Maintainer: Chenfeng Huang - [Kaggle](https://www.kaggle.com/alrickh)

For questions, please open an issue or discussion in this repository.

## License

Distributed under the terms of the license in `LICENSE`.

Kaggle_Silver_Medal_Solutioun_CMI-PIU

Was this content helpful?

Recent Feedback

Download README