IoT Machine Learning Gas Detection

This project is designed to be a learning tool to help educate myself and others on the practical application of machine learning on IoT data. Machine learning is used to classify what state a batch of time series data is in based on publicly available gas detection data.

This repo contains three parts:

1. A simulator to stream continuous data from a dataset over an MQTT broker as if it were real data coming from IoT sensors. With default run speed (50 rows per second) there is around 5.5 hours of data.
2. A Jupyter Notebook where an ML model was trained to predict what classification the sensors are currently emitting. The model was trained on the data from the simulator.
3. A python script that subscribes to the broker and utilizes the latest 480 points to run the model and publish a predicted state of the time series data.

Need to add MQTT broker credentials to the .env file in the following format. This setup will work by default on the HiveMQ publicly available broker. If you wish to use a private broker there are snippets of code in the mqtt_publish.py and run_model_live.py files that can be uncommented and used if you update the .env file accordingly.

MQTT_HOST="broker.hivemq.com"
MQTT_PORT=1883
MQTT_USERNAME=""
MQTT_PASSWORD=""
MQTT_TOPIC_PREFIX="LCOI_MOX_ML_IOT"

Running the simulator

1. Create a virtual environment and install the requirements from the requirements.txt file.
2. Run mqtt_publish.py

Running the prediction client

1. Create a virtual environment and install the requirements from the requirements.txt file.
2. Run run_model_live.py

Training a new model

1. Data can be downloaded from this link and placed in a folder called data in this dir.
2. Create a virtual environment and install the requirements from the requirements.txt file.
3. Run jupyter notebook and open Model Development.ipynb then run all cells. Note that the Arsenal kernel function will take a long time to run.