bareethul/AutoML-books-classification

Model Card for AutoML Books Classification

This model card documents the AutoML Books Classification model trained with AutoGluon AutoML on a classmate’s dataset of fiction and nonfiction books.
The task is to predict whether a book is recommended to everyone based on tabular features.

---

Model Details

• Developed by: Bareethul Kader
• Framework: AutoGluon
• Repository: bareethul/AutoML-books-classification
• License: CC BY 4.0

---

Intended Use

Direct Use

• Educational demonstration of AutoML on a small tabular dataset.
• Comparison of multiple classical ML models through automated search.
• Understanding validation vs. test performance tradeoffs.

Out of Scope Use

• Not designed for production or book recommendation engines.
• Dataset too small to generalize beyond classroom context.

---

Dataset

• Source: https://huggingface.co/datasets/jennifee/HW1-tabular-dataset .
• Task: Classification (RecommendToEveryone = 0/1).
• Size: 30 original samples + ~300 augmented rows.
• Features:
  • Pages (integer)
  • Thickness (float)
  • ReadStatus (categorical: read/started/not read)
  • Genre (categorical: fiction/nonfiction)
  • RecommendToEveryone (binary target)

---

Training Setup

• AutoML framework: AutoGluon TabularPredictor
• Evaluation metric: Accuracy
• Budget: 300 seconds training time, small scale search
• Hardware: Google Colab (T4 GPU not required, CPU sufficient)
• Search Space:
  • Tree based models: LightGBM, XGBoost, ExtraTrees, RandomForest
  • Neural nets: Torch, FastAI (small MLPs)
  • Bagging and ensembling across layers (L1, L2, L3)

---

Results

Mini Leaderboard (Top 3 Models)

Rank	Model	Test Accuracy	Validation Accuracy
1	RandomForestEntr_BAG_L1	0.55	0.65
2	LightGBM_r96_BAG_L2	0.53	0.72
3	LightGBMLarge_BAG_L2	0.53	0.74

• Best model (AutoGluon selected): RandomForestEntr_BAG_L1
• Test Accuracy: ~0.55
• Validation Accuracy (best across runs): up to 0.75 (LightGBM variants)

Note: The “best model” may vary depending on random splits and seeds.
While AutoGluon reported RandomForestEntr_BAG_L1 as best in this run, LightGBM models sometimes achieved higher validation accuracy but generalized less strongly.

---

Limitations, Biases, and Ethical Notes

• Small dataset size → models may overfit, performance metrics unstable.
• Augmented data → synthetic rows may not reflect true variability.
• Task scope → purely educational, not for real world recommendation.

---

AI Usage Disclosure

• ChatGPT (GPT-5) assisted in:
  • Helping with coding and AutoGluon AutoML approach on the go
  • Polishing the Colab notebook for clarity
  • Refining this model card

---

Citation

BibTeX:

@model{bareethul_books_classification,
  author       = {Kader, Bareethul},
  title        = {AutoML Books Classification},
  year         = {2025},
  framework    = {AutoGluon},
  repository   = {https://huggingface.co/bareethul/AutoML-books-classification}
}