Water Usage Signature Detection - ML Plan

This project is designed to add a machine learning feature for an iOS App that is paired with an IoT device that monitors water usage in a household. The IoT device stores high-resolution data in a time series database through InfluxDB integration, featuring real-time data visualization and usage tracking.

🌟 Goal

Train a supervised machine learning model to classify household water usage events by appliance or fixture (e.g., shower, faucet, toilet, washing machine, dishwasher, fridge, sprinkler).

---

🔍 Analyze Events & Engineer Features

🛠️ Existing Features:

• eventVolume
• avgFlowRate
• eventLength
• eventPeaks
• peakFlowRate
• timestamp

💡 Feature Engineering Ideas:

• ⏰ Time of Day: Extract hour, morning/afternoon/evening labels
• 📅 Day of Week: May help identify routine behaviors (e.g., laundry on weekends)
• ⏳ Time Since Last Event: Useful for session grouping and burst detection
• 🏖️ Is Weekend: Boolean feature to separate workday from rest day patterns
• 💥 Burst Indicator: Label short bursts (for ice maker, toilet flush)
• 📈 Rolling Stats: Moving average of volume or flow rate across last N events
• ➡️ Gap to Next Event: For future session detection or segmentation
• 🔗 Event Group ID: To tag related events (optional during pre-processing)

---

🔗 Group Related Events into Sessions (optional)

Used for appliances with multiple bursts (e.g., washing machine, dishwasher).

📜 Rules:

• Group events within 30 min of each other
• Total session duration, volume, flow, and event count
• Label the grouped session when patterns match known appliance usage

---

🏷️ Label the Dataset for Training

Start by applying rule-based heuristics to label ~200–300 samples manually.

Condition	Label
eventLength > 300, eventVolume > 10	shower
eventLength < 10, eventVolume < 1	faucet or toilet
Multiple bursts over 30 min, 5–10 gal total	washing_machine
Single long event, 3–6 gal, >20 min	dishwasher
Very short, low volume (<0.2 gal)	fridge/ice
Long duration, high volume (>50 gal)	sprinkler

---

🤖 Train the Classifier

🧪 Models to Try:

• Random Forest
• kNN
• Support Vector Machine (SVM)

🛠️ Pipeline:

• Train/test split
• Feature scaling
• Cross-validation
• Accuracy, F1-score, confusion matrix

---

📊 Evaluate and Iterate

• Analyze model performance
• Identify and refine poorly predicted labels
• Revisit labeling rules and session grouping
• Add more labeled data

---

📱 Deploy to Mobile (TensorFlow Lite)

• Convert trained model to TFLite
• Optimize with quantization
• Embed in mobile app (Android/iOS)
• Run inference locally for real-time classification

---

🐍 Python Script Overview

The feature_engineering.py script is an interactive tool for preprocessing and feature engineering of water usage datasets. It allows users to:

• Add time-based features (e.g., hour, day of the week, part of the day).
• Group events into sessions based on a time threshold.
• Add burst indicators for specific event patterns.
• Apply preset rules from a JSON file.
• Label the dataset interactively based on user-defined conditions.
• Remove rows without labels or specific burst indicators.
• Save the processed dataset to a new file.

---

🚀 How to Run the Script

To use the dataPreparation.py script for preprocessing and feature engineering, follow these steps:

1. Install Dependencies: Ensure you have Python 3.8 or later installed. Install the required libraries using pip:

   pip install pandas

2. Prepare Your Dataset: Place your CSV file(s) in the same directory as the script or specify the path when prompted.

3. Run the Script: Execute the script in your terminal:

   python3 dataPreparation.py

4. Follow the Interactive Prompts:

   • Select the dataset file to process.
   • Choose whether to add time-based features, group events into sessions, or add burst indicators.
   • Apply preset rules or label the dataset interactively.
   • Save the processed dataset to a new file or overwrite the existing one.

5. Output: The processed dataset will be saved in the specified output file, ready for further analysis or model training.