Smart Home Energy Predictions: IoT and Machine Learning

AAI-530 IoT and Machine Learning Final Project

University of San Diego, School of Engineering - Masters of Applied Artificial Intelligence

Project Overview

This project leverages IoT sensor data to develop predictive models and classifications for indoor environmental conditions. Using deep learning techniques, including Long Short-Term Memory (LSTM) networks, Convolutional Neural Networks (CNN), and Seasonal Autoregressive Integrated Moving Average with Exogenous Variables (SARIMAX), the study explores temperature forecasting and anomaly detection in smart home settings.

Project Components

Final Paper

The technical paper provides an in-depth analysis of the project, detailing data preprocessing, exploratory data analysis (EDA), model selection, and performance evaluation.

Key insights include:

• Data sources: EcoLab Ground and WeatherLink Indoor sensors.

• Data preprocessing: Outlier removal, missing value handling, feature engineering.

Modeling results:

• LSTM achieved a Root Mean Squared Error (RMSE) of 0.81°C.

• CNN classified indoor conditions with 94% accuracy but struggled with rare classes.

• SARIMAX performed well for short-term forecasts but declined in long-term accuracy.

Dashboard

[View Tableau - Smart Home IoT Dashboard]

The interactive dashboard visualizes key findings, including:

• Time-series analysis of temperature trends from June to November 2023.

• Comparison of predicted vs. actual temperatures using LSTM and SARIMAX.

• eCO₂ levels and classification results for different environmental conditions.

Data Files

• The dataset contains cleaned IoT sensor data used for modeling and analysis:

• EcoLab Ground Cleaned.csv – Cleaned dataset from the EcoLab Ground sensor.

• Front Door Cleaned.csv – Processed data from the Front Door sensor.

• Weather Link Indoor Cleaned.csv – Indoor environmental sensor data.

• Model predictions

• Model performance metrics

• Visuals

Trained Models

The following trained models are included:

• final_cnn_model.keras – Final trained CNN model.

• lstm_model_optimized.keras – Optimized LSTM model.

Codebase

The code repository consists of Python scripts and Jupyter Notebooks for:

• Data preprocessing: Cleaning and feature engineering.

• Modeling: Implementation of LSTM, CNN, and SARIMAX models.

• Evaluation: Performance metrics and comparative analysis.

• Visualization: Graphs and plots illustrating trends and predictions.

Git Ignore & License

.gitignore specifies ignored files, ensuring version control efficiency.

LICENSE outlines usage rights, adhering to Apache License 2.0 for software and Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) for dataset use.

How to Use

1. Clone the repository:

git clone https://github.com/oxayavongsa/aai-530-iot-smart-house.git

2. Install dependencies:

pip install -r requirements.txt

3. Run the Jupyter Notebook:

• Open and execute Final_Code_G3.ipynb in Jupyter or Google Colab.

4. Explore results:

• View predictions and insights in the provided dashboard.

Key Findings

• LSTM outperformed traditional time-series models in capturing temperature variations.

• CNN effectively classified environmental conditions but required better handling of imbalanced classes.

• SARIMAX provided interpretability but struggled with long-term forecasting.

• Data preprocessing, including smoothing and interpolation, significantly improved model performance.

Future Enhancements

• Incorporate real-time IoT data streams for live monitoring.

• Explore hybrid models combining CNN and LSTM for enhanced spatial and temporal analysis.

• Improve handling of rare conditions through class balancing techniques.

Contributors

• Outhai Xayavongsa (Ms. Thai) - Team Leader

• Aaron Ramirez - Tech Lead

License

• The software is licensed under Apache License 2.0.

• The dataset follows the Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license.