AbdullahImran/Mental-Health-Classification-Dataset

Mental Health Classification Dataset

A curated English text dataset for binary mental-health text classification.

The dataset contains two classes:

• —MH — mental-health-related content
• —Non-MH — non-mental-health-related content

The dataset was prepared as part of a broader NLP project involving data cleaning, deduplication, edge-case handling, SBERT fine-tuning, and downstream binary classification.

Dataset Summary

The dataset is provided as two CSV files:

data/
├── mental_health.csv
└── non_mental_health.csv

Data Fields

Mental Health Dataset

The mental_health.csv file contains:

For the mental-health dataset, is_mental_health = True represents mental-health-related content.

Non-Mental Health Dataset

The non_mental_health.csv file contains:

For the non-mental-health dataset, is_mental_health = False represents non-mental-health content.

Data Preparation

The final dataset used for this repository was produced through a preprocessing pipeline that included:

1. 1.Data cleaning
2. 2.Duplicate removal
3. 3.Dataset consolidation
4. 4.Label verification
5. 5.Edge-case identification
6. 6.Separation of mental-health and non-mental-health examples

The repository contains the final curated CSV files rather than intermediate dataset-generation versions.

Example

A mental-health example may contain text discussing a mental-health condition, symptoms, treatment, or related experiences.

A non-mental-health example may discuss topics such as technology, consumer products, entertainment, science, or other general subjects without representing mental-health content.

Intended Use

This dataset is intended for:

• —Natural language processing research
• —Binary text classification
• —Mental-health-related NLP experimentation
• —Sentence embedding research
• —Training and evaluating machine-learning classifiers
• —Educational and research projects

It can be used to investigate whether a piece of text contains mental-health-related content.

Relationship to the Model

This dataset was used as part of the broader project associated with the Fine-Tuned SBERT MiniLM for Mental Health Classification model.

The classification pipeline uses a fine-tuned SBERT encoder to generate sentence embeddings, followed by a Logistic Regression classifier for binary classification.

Text
  |
  v
Fine-Tuned SBERT
  |
  v
384-Dimensional Embedding
  |
  v
Logistic Regression
  |
  v
MH / Non-MH

Limitations

The dataset should not be interpreted as a clinical or diagnostic resource.

Mental-health language is complex and highly contextual. A text classifier trained on this dataset may produce incorrect classifications, particularly for:

• —Ambiguous statements
• —Indirect references to mental health
• —Sarcasm or figurative language
• —Unfamiliar writing styles
• —Topics outside the dataset distribution
• —Culturally or linguistically different expressions

Dataset labels represent the classification scheme used during dataset preparation and should not be interpreted as clinical judgments.

Responsible Use

This dataset should be used responsibly.

A prediction that text is classified as MH does not indicate that the author of the text has a mental-health condition. Similarly, a Non-MH prediction does not establish the absence of mental-health concerns.

The dataset and models derived from it should not be used as a substitute for professional medical or psychological assessment.

Citation

If you use this dataset in a project or publication, please cite this Hugging Face repository and the associated project.

License

No explicit open-source license is specified for this dataset at this time. Users should review the provenance and intended usage of the underlying data before redistributing or using it commercially.