An original data science metric that quantifies smartphone behavioral intensity as a single score between 0 and 1.
Developed by M Sarvesh · Data Science Research, 2026
Most screen time tools tell you how long you used your phone. DUII tells you how intensely — by measuring both your behavioral dependency and how much your usage is disrupting your daily routine.
DUII combines two original sub-indices derived from correlation analysis on four real-world datasets:
| Index | What it measures |
|---|---|
| BDI (Behavioral Dependency Index) | How addicted your usage patterns look |
| RDI (Routine Disruption Index) | How much your phone is disrupting sleep & routine |
DUII = 0.43 × BDI + 0.57 × RDI
BDI = 0.70 × U + 0.29 × C + 0.01 × B
RDI = 0.52 × U + 0.48 × (1 − S_normalized)
Where:
- U = normalized daily usage →
(x − 0) / (11.5 − 0) - C = normalized phone checks →
(x − 20) / (150 − 20) - B = normalized screen before bed →
(x − 0) / (2.6 − 0) - S = normalized sleep hours →
(x − 3.8) / (9.6 − 3.8)
Every weight (0.43, 0.57, 0.70, 0.29, 0.52, 0.48) was derived through correlation analysis — not assumed.
| Score | Category | Meaning |
|---|---|---|
| 0.00 – 0.30 | 🟢 Low | Healthy digital habits |
| 0.30 – 0.60 | 🟡 Moderate | High but controlled |
| 0.60 – 0.80 | 🟠 High | Risk of dependency |
| 0.80 – 1.00 | 🔴 Severe | Strong digital intensity |
A full Android app built in React Native around the DUII formula.
- Auto Fetch — reads real screen time from Android's UsageStats API
- Manual Input — sliders for all 4 variables with live score calculation
- History Tracking — saves every entry locally on device with charts
- Insights Tab — trend detection, 7-day projection, early burnout alerts
- Dark & Light mode
- 100% local — no data sent to any server
The Insights tab runs statistical analysis on your saved history:
- Linear regression to detect if your score is trending up or down
- 3-day moving average to smooth daily noise and surface the real trend
- 7-day linear projection to show where your score is heading
- Smart alerts that fire when trend exceeds thresholds (e.g. rising >1%/day, 3+ consecutive days in High zone)
- Variable impact ranking — tells you which of the 4 inputs is hurting your score most
| Step | What was done |
|---|---|
| Data Collection | 4 real-world datasets from Kaggle covering smartphone usage, app engagement, behavioral addiction, sleep patterns |
| Cleaning | Removed nulls and duplicates across all datasets |
| Feature Selection | Split columns into Core (used for DUII) and Validation (used only for verification) |
| EDA | Distribution plots, correlation heatmaps, outlier detection, usage vs addiction scatter plots |
| Feature Engineering | Derived social_media_ratio, usage_ratio, sleep deficit columns |
| Normalization | Min-Max scaling to keep all variables in [0, 1] range |
| BDI Construction | Weights derived from correlation with addiction-related outcomes |
| RDI Construction | Weights derived from correlation with sleep and academic disruption |
| Validation | Both indices validated against held-back validation columns |
| DUII Construction | α=0.43, β=0.57 found through optimization of predictive importance |
- Node.js v18+
- Java JDK 17
- Android Studio + Android SDK
- Android device or emulator (API 28+)
# Clone the repo
git clone https://github.com/yourusername/duii-app.git
cd duii-app
# Install dependencies
npm install
# Start Metro bundler
npx react-native start
# Run on Android (in a second terminal)
npx react-native run-androidOn first launch, the app will ask for Usage Access permission to read real screen time data. Go to Settings → Usage Access → enable DUIIApp.
duii-app/
├── App.tsx # Main app — all screens and logic
├── android/
│ └── app/src/main/java/com/duiiapp/
│ ├── UsageStatsModule.java # Native Android module for real data
│ ├── UsageStatsPackage.java # Package registration
│ └── MainApplication.kt # App entry point
├── research/
│ └── DIGITAL_USAGE_INTENSITY_INDEX.docx # Full research document
└── README.md
- Google Fit integration for real sleep hours
- Push notifications for daily DUII score
- Phase 2 — ARIMA/LSTM model via Python backend for true ML forecasting
- CSV export for research data collection
- Multi-user comparison (anonymized)
- Potential OS-level integration (Android Digital Wellbeing / Apple Screen Time)
M Sarvesh Data Science Research Study, 2026
This project and the DUII formula are original research work by M Sarvesh. Feel free to reference with attribution.