Mental Health & Social Media Analysis

Executive Summary

Social media usage has increased significantly in recent years, raising concerns about its potential impact on mental health and productivity. Understanding this relationship is essential for organizations and indivisuals aiming to improve well-being and performance.

In this project, I analyzed social media usage patterns ans their relationship with stress, anxiety, and productivity indicators. The analysis involved data cleaning, exploratory analysis, visualization, and machine learning modeling.

To explore predictive patterns, I developed two Random Forest models and applied clustering techniques (KMeans). The first model, based on original stress values, achieved an accuracy close to 50%, suggesting that the relationship between social media usage and stress is complex and not easily predictable.

However, after simplifying the problem by binarizing stress and productivity variables, the second model achieved approximately 80% accuracy. This highlighted the importance of problem formulation and feature engineering in machine learning projects.

The results indicate that while social media usage does not show a strong direct relationship with stress and anxiety, a clearer relationship was observed with productivity levels.

This project demonstrates how data analysis and machine learning can help uncover patterns in complex behavioral data and support data-driven decision-making.

Key Business Insights

• No strong relationship between social media usage and stress/anxiety

• Clearer relationship between social media usage and productivity

• Model performance improved from ~50% to ~80% after target simplification

• Behavorial patterns identified through clustering

Tools & Technologies

• Python

• Pandas & NumPy

• Scikit-Learn

• Power BI

• Jupyter Notebook

Model Performance

Two Random Forest models were developed to explore predictive patterns:

• Model 1 (original stress values): ~50% accuracy, indicating no clear predictive relationship

• Model 2 (binarized stress and productivity): ~80% accuracy, improving predictive performance

These results highlight the complexity of predicting mental health indicators and the importance of problem formulation and feature engineering.

Next Steps

• Collect additional behavorial and demographic data

• Explore more advanced Machine Learning models

• Perform deeper segmentation analysis

• Validate findings with real-world datasets