Mental Health Score Predictor

A regression model that predicts a person's Mental Health Score (0–10) from their social media usage habits, sleep, study, physical activity, and stress level.

The goal is simple: understand how daily digital habits and lifestyle factors relate to mental well-being, and offer a working baseline model others can build on.

Model Details

Model type: Random Forest Regressor (scikit-learn)
Task: Tabular regression
Target variable: Mental_Health_Score (continuous, 0–10)
Framework: scikit-learn, trained inside a Pipeline
Saved format: .pkl (via joblib)

Intended Use

Educational / portfolio project exploring the link between digital habits and mental well-being.
Baseline for further experimentation (feature engineering, other algorithms, hyperparameter tuning).

Not intended for: clinical, diagnostic, or any real-world mental health decision-making. This model is trained on a small, self-reported survey-style dataset and does not represent a validated psychological assessment.

Dataset

Size: 5,000 rows, 13 columns
Features used:
- Age, Gender, Country, Academic_Level
- Most_Used_Platform, Purpose_Of_Use
- Avg_Daily_Usage_Hours, Daily_Unlocks
- Study_Hours, Physical_Activity_Hours, Sleep_Hours_Per_Night
- Stress_Level (ordinal: Low → Very High)
Target: Mental_Health_Score
No missing values in the raw data.

Preprocessing

Duplicates removed; Physical_Activity_Hours clipped at 0 (no negative values).
Country grouped into top-10 countries + "Other" to reduce cardinality.
Study_Hours (right-skewed) log-transformed, then scaled.
Other numeric features standard-scaled.
Stress_Level ordinal-encoded (Low → Very High).
Remaining categorical features one-hot encoded.
80/20 train-test split.

Training & Evaluation

Four models were compared, with Random Forest tuned via Grid Search and Randomized Search:

Model	R² (Test)	R² (Train)	MAE	MSE	RMSE
Linear Regression	0.740	0.724	0.536	0.457	0.676
Random Forest (default)	0.878	0.981	0.347	0.215	0.464
Random Forest (Grid Search)	0.872	0.969	0.358	0.225	0.474
Random Forest (Random Search)	0.865	0.955	0.369	0.237	0.487

The default Random Forest performed best on the test set and is the model saved (Mental_Health_Model.pkl). Tuned versions traded a bit of test performance for lower variance — worth exploring further if generalization to new data is a priority.

How to Use

import joblib
import pandas as pd

model = joblib.load("Mental_Health_Model.pkl")

sample = pd.DataFrame([{
    "Age": 21,
    "Gender": "Male",
    "Academic_Level": "Undergraduate",
    "Most_Used_Platform": "Instagram",
    "Purpose_Of_Use": "Entertainment",
    "Avg_Daily_Usage_Hours": 4.5,
    "Daily_Unlocks": 140,
    "Study_Hours": 4.0,
    "Physical_Activity_Hours": 2.0,
    "Sleep_Hours_Per_Night": 6.5,
    "Stress_Level": "Medium",
    "Grouped_Country": "Other"
}])

prediction = model.predict(sample)
print(f"Predicted Mental Health Score: {prediction[0]:.2f}")

Note: the model expects the same preprocessing pipeline it was trained with (it's bundled inside the saved Pipeline), so raw feature columns can be passed directly as shown above.

Limitations

Trained on a single, likely synthetic/survey-style dataset — may not generalize to real-world populations.
Self-reported features (usage hours, stress level) are prone to bias.
Correlation-based insights should not be read as causation.
No fairness/bias audit across demographic groups has been performed.

Files

notebook.ipynb — full workflow: EDA, cleaning, feature engineering, training, tuning, and evaluation.
data.csv — training dataset.
Mental_Health_Model.pkl — trained Random Forest pipeline.

License

Released under the MIT License. Use freely for learning and experimentation.

Downloads last month: -; Downloads are not tracked for this model. How to track