srivathsanb14/indoor-outdoor-text

Indoor Outdoor Text Dataset

Dataset Summary

This dataset contains short English descriptions of places and everyday settings. Each description has a binary label: indoor or outdoor.

Purpose

The dataset was created for a course assignment on building and augmenting a text dataset. It is intended for demonstrating binary text classification and label-preserving text augmentation.

Composition

Each record contains:

• —text: a short English description.
• —label: either indoor or outdoor.

The original collection contains 103 balanced examples. These originals were split (~70/15/15, stratified by label) into training, validation, and test portions before augmentation. Synthetic examples were generated only from the training originals, so validation and test remain unaugmented.

Data Collection

The original examples describe ordinary places and activities such as campus buildings, restaurants, museums, stores, parks, trails, plazas, markets, and outdoor recreation areas.

Preprocessing and Augmentation

Original texts are checked for duplicate entries and valid binary labels, then split into train/validation/test sets stratified by label.

Synthetic examples are generated from the training originals only, using controlled, label-preserving transformations such as word substitutions, word insertions/removals, word swaps, and sentence-level prefix/suffix variations. The original label is retained for every augmented example. A fixed random seed is used for reproducibility, and exact duplicate texts are rejected.

Labels

• —indoor: the described setting/activity primarily occurs inside a building or enclosed space.
• —outdoor: the described setting/activity primarily occurs outside.

Splits

• —train: training originals plus their augmented versions.
• —validation: unaugmented validation originals (~15% of the original collection).
• —test: unaugmented test originals (~15% of the original collection).

Intended Use

Educational experimentation with binary text classification, dataset construction, data augmentation, and reproducible machine-learning workflows.

Limitations

The dataset is relatively small and focuses on English descriptions of places and settings. The labels represent the primary setting described in each text and may not capture mixed indoor/outdoor situations. The descriptions are not a comprehensive representation of any city or society.

Ethical Considerations

The dataset describes ordinary places and activities and is not intended to identify or profile individuals. It does not intentionally include private personal information. Users should avoid adding personally identifying or sensitive information.

License

CC BY 4.0

AI Usage Disclosure

Generative AI (Claude) was used during development to assist with drafting/revising example descriptions and designing the augmentation workflow.