Model Card for keerthikoganti/sleep-hours-predictor-automl-ag

This dataset contains self-reported technology-use patterns and routines.

Model Details

Model Description

This dataset contains self-reported technology-use patterns and routines, plus sleep timing/quality attributes. It was created as a class exercise for supervised learning on tabular data. The primary task is regression — predicting nightly sleep_hours. The same features can also support simple classification tasks if desired.

• Developed by: Keerthi Koganti
• Model type: AutoML (AutoGluon Tabular ensemble; model family chosen via search)
• Language(s) (NLP): English
• Task: Tabular Regression
• Target column: sleep_hours (continuous, hours slept per night)
• License: Carnegie Mellon educational use
• Framework: autogluon.tabular
• Repo artifacts: autogluon_sleep_model.zip (zipped native AutoGluon predictor directory) metrics.json (test-set metrics for reproducibility)

Model Sources

• Repository: Iris314/Students_sleep_tabular

Uses

Direct Use

Classroom demos of AutoML on tabular regression tasks

Baseline experiments for feature engineering and evaluation

Comparing different AutoML presets and model search spaces

Out-of-Scope Use

Production deployment or any sleep/health recommendation system

Generalization beyond course context and small cohort data

Clinical or safety-critical applications

Bias, Risks, and Limitations

Small sample size and potential sampling bias.

Self-report bias in device use and sleep estimates.

Domain shift likely for other age groups, locations, or lifestyles.

Recommendations

Use primarily for teaching and demonstration of tabular ML workflows. If you publish results, disclose the split strategy, preprocessing, and any imputations/encodings performed by AutoGluon.

How to Get Started with the Model

Use the code below to get started with the model.

import pathlib, shutil, zipfile import huggingface_hub as hf from autogluon.tabular import TabularPredictor

REPO = "keerthikoganti/sleep-hours-predictor-automl-ag" ZIPNAME = "autogluon_sleep_model.zip"

dest = pathlib.Path("hf_download") dest.mkdir(exist_ok=True)

Download zipped predictor directory

zip_path = hf.hf_hub_download( repo_id=REPO, filename=ZIPNAME, repo_type="model", local_dir=str(dest), local_dir_use_symlinks=False, )

Extract to folder

extract_dir = dest / "predictor_dir" if extract_dir.exists(): shutil.rmtree(extract_dir) extract_dir.mkdir(parents=True, exist_ok=True)

with zipfile.ZipFile(zip_path, "r") as zf: zf.extractall(str(extract_dir))

Load predictor from native directory

predictor = TabularPredictor.load(str(extract_dir))

Example: predictions on a new DataFrame X

preds = predictor.predict(X)

Training Details

Training Data

Dataset: Iris314/Students_sleep_tabular

Splits: 80/20 random split (random_state=42) on the augmented data (≈300 rows)

Target column: sleep_hours

Training Procedure

Library: AutoGluon Tabular

Presets: "best_quality" (ensembles across boosted trees, RF, kNN, neural nets, etc.)

Training time limit: 300 seconds

Evaluation metric (internal): AutoGluon default for regression (root_mean_squared_error)

Training Hyperparameters

• Training regime: time_limit: 300s

presets: "best_quality"

random_state: 42

problem_type: inferred automatically by AutoGluon

eval_metric: AutoGluon default for regression

Evaluation

Testing Data

Held-out 20% of the augmented split (≈60 rows)

Metrics (replace with actuals from metrics.json)

R² = 0.78

MAE = 0.62 hours

RMSE = 0.95 hours

Environmental Impact

Training was lightweight and classroom-focused:

Hardware: CPU runtime in Google Colab (no GPU)

Training wall-time: ≤ 5 minutes

Estimated emissions: negligible

Cloud provider: Colab (Google)

Model Card Contact

Keerthi Koganti — kkoganti@andrew.cmu.edu