Device Storage Growth Forecaster

A machine learning portfolio project that forecasts laptop/phone storage usage over the next 30, 60, and 90 days.

Project Goal

Predict how end-user device storage fills over time, then compare forecasting approaches such as:

• Baseline trend models
• Prophet
• XGBoost with lag features
• Optional LSTM (stretch goal)

Project Structure

storage_forecaster_v2/
├── configs/                  # Config files
├── dashboard/                # Streamlit app
├── data/
│   ├── raw/                  # Real raw device logs (future)
│   ├── processed/            # Cleaned/model-ready datasets
│   └── synthetic/            # Synthetic training data
├── models/                   # Saved trained models
├── notebooks/                # EDA and experimentation notebooks
├── reports/
│   └── figures/              # Exported plots for README/report
├── scripts/                  # CLI scripts for data generation/training
├── src/
│   ├── data/                 # Data loading and preprocessing
│   ├── evaluation/           # Metrics and backtesting
│   ├── features/             # Feature engineering
│   ├── models/               # Model training/inference code
│   └── visualization/        # Plotting utilities
└── tests/                    # Unit tests

Quick Start on Windows (VS Code + venv)

1) Open terminal in the project folder

Use Command Prompt or the VS Code terminal.

2) Create a virtual environment

python -m venv .venv

3) Activate it

.venv\Scripts\activate

4) Install dependencies

pip install --upgrade pip
pip install -r requirements.txt

5) Run the synthetic data generator

python scripts\generate_synthetic_data.py

6) Run the visualization script

python scripts\visualize_data.py

7) Launch the dashboard! 🚀

streamlit run dashboard\app.py

Using the Dashboard

Test with Sample Data

The dashboard comes with synthetic sample data - just run it and explore!

Test with Your Own Data 📋

New in the dashboard: Data Builder tab makes it super easy!

1. Open the dashboard

2. Click the "Data Builder" tab

3. Choose one of three methods:

   • 📊 See Example Data - View a real example (no download)
   • 🔨 Generate Template - Create a blank CSV template in 30 seconds
   • 📥 Download Real Example - Get sample data to test immediately

4. For your own data:

   • Use "Generate Template" to create a pre-formatted CSV
   • Fill in your device's storage data (7-30 days minimum)
   • Upload to Forecast Lab tab
   • Get predictions! 🎉

See DATA_BUILDER_GUIDE.md for detailed instructions.

Recommended VS Code Setup

Install these extensions:

• Python
• Pylance
• Jupyter

Then in VS Code:

1. Open the folder
2. Press Ctrl+Shift+P
3. Select Python: Select Interpreter
4. Choose the interpreter inside .venv

First Milestones

• Create synthetic dataset
• Visualize storage behavior by profile
• Build EDA notebook
• Train baseline forecasting model
• Train Prophet model
• Train XGBoost model
• Evaluate 30/60/90-day forecasts
• Build upgraded Streamlit dashboard

Synthetic User Profiles

• Media-heavy: steady photo/video growth
• Gamer: occasional large app/game installs
• Office user: slow, linear document-heavy growth
• Cleaner: periodic cleanup cycles

Example Commands

Train simple baseline model

python scripts\train_baseline.py

Train improved baseline (multi-horizon Ridge with lag features)

python scripts\train_baseline_v2.py

Train Prophet forecasts

python scripts\train_prophet.py

Train XGBoost forecasts

python scripts\train_xgboost.py

Initialize the review store

python scripts\init_review_store.py

Export reviews to CSV

python scripts\export_reviews.py

Compare model metrics

python scripts\compare_model_metrics.py

Launch dashboard

streamlit run dashboard\app.py

Notes

• The synthetic dataset is intended for prototyping and portfolio work.
• You can later add a real device logging script for Windows to collect actual usage data.
• Use time-based validation instead of random train/test splits.
• Reviews can use local SQLite in development or managed PostgreSQL in deployment via DATABASE_URL.
• See DEPLOYMENT.md, PRODUCTION_ARCHITECTURE.md, and .env.example before public launch.

Current recommended production path

1. Train or refresh models.
2. Configure a managed PostgreSQL database.
3. Set DATABASE_URL in your deployment secrets.
4. Deploy the Streamlit app via Community Cloud or Docker.
5. Collect real user reviews and forecast logs for iteration.