<aside> 🗨️

Goal of this analysis

The goal of this analysis is to

1. Analyze and compare content metrics, posting habits, engagement rates, and other behavioral attributes of high-growth daily posters versus lower-growth daily posters.
2. Evaluate whether certain types of content perform better when posted daily, and under what conditions.
3. Use predictive modeling to estimate creator growth based on posting behavior and content features and compare model predictions against actual performance metrics. </aside>

<aside> 🗨️

Deliverables

Python-based Jupiter Notebook analyzing:

• Creator posting frequency
• Engagement and follower growth metrics
• Daily vs non-daily behavior patterns
• Tools:
  • Machine learning models
  • Actual vs predicted comparison for validation and insight generation

Behavioral Segmentation:

• Clustering or comparative analysis of high-growth daily posters vs low-growth peers </aside>

<aside> 🗨️

Questions to be answered

Objective:

Identify which health and social indicators (beyond depression alone) most influence changes in depression rates across countries and years.

Machine Learning Framing:

• Analyze feature importances from Random Forest values to identify the most impactful predictors.
• Segment data by year, region, or development level to find differentiating traits in high-depression vs. low-depression populations.
• Use clustering or correlation matrices to group related conditions and explore co-occurring patterns.
• Actual vs Predicted Depression Prevalence, this is to test model accuracy of the machine learning model can predict real world depression prevalence. </aside>

<aside> 🗨️

What data do I have?

• Source:
  • Kaggle – Mental Health Dataset by programmerrdai
  • Structured tabular data (CSV format)
• Characteristics
  • Size: ~1,000–2,000 records depending on filtering based on different countries
  • Missing Data:
    • Some columns have NaN values (e.g., due to country-specific availability)
    • Columns cleaned via:
      • Dropping overly sparse features
      • Imputing missing numeric data using column mean
      • Removing irregular numbers </aside>

<aside> 🗨️

Model Evaluation Metrics

Linear Regression performs exceptionally well and provides insights; it may not capture the complex linear relationship of the data but through the values.

Both models perform well but Random Forest is more accurate the small performance gap shows it is strong linearly correlated with predictors, but Random Forest still add precision by capturing non-linearities and interactions.

The high accuracy suggests your dataset is clean, well-structured and highly predictive of the target, this shows that the dataset is a solid a dwell structured dataset and may be used for real world health predictions.

</aside>

<aside> 🗨️

Visualization from the graph

• The model has achieved excellent predictive performance, as shown in the graph there's a near perfect straight pattern that confirms there's a strong linear correlation between actual values and predicted values.
• This proves that my model is visually and statistically validated to be use in predicting depression prevalence in real world policy or different health surveillance scenarios. </aside>

<aside> 🗨️

Conclusion

This project successfully applied supervised machine learning techniques to predict depression prevalence using a global mental health dataset. Moreover, Im also able to utilize Linear Regression and Forest Regressor based on their ability to accurately predict depression rates.

• Both models performed exceptionally well, with Random Forest achieving the highest accuracy.
• Both models performed well but as shown above Random Forest achieving highest accuracy, the linear correlation as shown in the graph above the actual values and predicted values has nearly perfected regression plot reinforcing the accuracy of the model’s prediction.
• Limitations in the project:
  • The data set is based on county level statistics not individual level data this limits the actual accuracy of measuring depression.
  • Historical prevalence rates lack real time updates which would affect the accuracy of the dataset as well.
  • There are several missing variables and values. </aside>

Refference

Dataset: kaggle.com/datasets/programmerrdai/mental-health-dataset

https://github.com/Ethan004-Code/Mental_Health-Research-