![[44afc8_5d4e430a4a0a40ee8454bcbe1efb5551~mv2.jpg]] Smartphones have become deeply integrated into our daily lives, but not everyone develops unhealthy usage patterns. While age, gender, and gaming habits are often discussed as potential risk factors, I wanted to answer a different question: > **Can smartphone addiction be accurately predicted from how people use their phones?** To explore this, I built an interpretable **Logistic Regression** model to identify the behavioral factors most strongly associated with smartphone addiction. # Dataset The dataset contains **691,369 users** with behavioral, demographic, and lifestyle information. ## Features | Category | Features | | ------------------------- | ------------------------------------------------------------------------ | | **Demographics** | Age, Gender | | **Phone Usage** | Daily Screen Time, Weekend Screen Time, Social Media Hours, Gaming Hours | | **Lifestyle** | Sleep Hours, Work/Study Hours | | **Phone Engagement** | Notifications Per Day, App Opens Per Day | | **Self-Reported Factors** | Stress Level, Academic/Work Impact | | **Target** | Smartphone Addiction (0 = No, 1 = Yes) | # Modeling Approach The modeling pipeline included: - Missing value imputation - Standardization of numerical variables - One-hot encoding for categorical variables - Logistic Regression with L2 regularization - 5-fold Stratified Cross Validation ## Model Performance | Metric | Score | | -------- | --------: | | Accuracy | **83.5%** | | ROC-AUC | **0.912** | | F1 Score | **0.887** | The nearly identical training and validation scores indicate excellent generalization with minimal overfitting. # Which Features Matter Most? After fitting the model, I examined the standardized coefficients and odds ratios. | Rank | Feature | Odds Ratio | | ---- | ------------------- | ---------: | | 1 | Social Media Hours | **3.40** | | 2 | Weekend Screen Time | **2.57** | | 3 | Daily Screen Time | **2.41** | | 4 | Sleep Hours | 1.10 | | 5 | App Opens Per Day | 1.10 | | 6 | Gaming Hours | **0.91** | # Key Findings ## 1. Smartphone Addiction Is Primarily a Behavioral Problem The model shows that **behavioral phone usage overwhelmingly outperforms demographic characteristics**. The strongest predictors are: - Time spent on social media - Weekend screen time - Overall daily screen time Meanwhile: - Age - Gender - Stress level - Academic/work impact contributed very little once actual phone usage behavior was included. This suggests that smartphone addiction is driven far more by **how people use their phones** than by **who they are**. ## 2. Social Media Dominates Smartphone Addiction The single strongest predictor was **daily social media usage**. After standardizing all numerical variables, the model estimates: > A one standard deviation increase in social media usage is associated with approximately **3.4× higher odds** of being classified as smartphone addicted. This makes social media usage by far the most influential behavioral signal in the dataset. ## 3. Weekend Usage Reveals Compulsive Behavior Weekend screen time was more predictive than overall daily screen time. One possible explanation is that weekdays naturally impose external constraints such as work or school. During weekends, users have greater control over how they spend their time. Individuals who continue spending excessive time on their phones despite having fewer external obligations may exhibit stronger habitual or compulsive usage patterns. ## 4. What Users Do Matters More Than How Long They Use Their Phones The three strongest predictors ranked as: ```text Social Media Hours > Weekend Screen Time > Daily Screen Time ``` This suggests that **the type of smartphone activity is more informative than total screen time alone**. Simply spending many hours on a smartphone does not necessarily indicate addiction. Instead, prolonged engagement with highly rewarding applications—particularly social media—appears to be a much stronger behavioral signal. ## 5. Gaming Hours Were Surprisingly Unimportant One of the most interesting findings was that **gaming hours contributed very little** once overall smartphone usage was considered. Its odds ratio was close to 1 (0.91), making it one of the weakest predictors in the model. This suggests that: - smartphone addiction is **not simply driven by gaming**, - excessive social media engagement appears to be much more indicative of problematic smartphone use, - gaming and smartphone addiction may represent related but distinct behavioral patterns. In other words, **heavy gamers are not necessarily smartphone addicted**, while heavy social media users are much more likely to exhibit addictive smartphone behaviors. # Final Thoughts This project highlights several important insights. - Smartphone addiction is better explained by **behavioral usage patterns** than demographic characteristics. - Social media usage is by far the strongest behavioral indicator. - Weekend usage captures habitual phone dependence beyond average daily usage. - Gaming contributes surprisingly little once broader smartphone behaviors are considered. - Simple, interpretable models such as Logistic Regression can achieve excellent predictive performance while providing meaningful behavioral insights.