Hello,
I have a few sensors in a home. I am going to continuously receive data fro these sensors, and do Machine Learning (in Python. Keras) on stored data every 24 hours. One possibility to do that is to use IoT platforms to collect all the data and do machine learning on a schedule. However, every IoT platform that I found is to be paid. I need to repeat this process for many buildings, so I am searching for alternative ways to be more cost-effective to be implemented.
What alternative ways do you suggest?
Thank you
What machine learning task you are interested to perform on your data? For the sensory data most researchers perform evaluation of optimal position and velocity or for application of object tracking or detection problems etc. For cleaning such data, you may need Kalman Filter, Butterworth Filter or Grassmann Filter etc.
There are no exactly best practices. But when you are do many projects; in the mean time you can come up with many best practices. And of course you can follow others to know more. For details about working with sensor data ceck out these links:
https://www.mathworks.com/videos/cleaning-and-analyzing-real-world-sensor-data-with-matlab-1568991034921.html
https://www.youtube.com/watch?v=zC0NcIfPgqU
https://www.youtube.com/watch?v=YnSFhluG5Cw
https://www.youtube.com/watch?v=g8MfWGibjT8
https://www.youtube.com/watch?v=6K6OZ2APbKk
https://www.youtube.com/watch?v=SH5nlNY5cO4
https://www.youtube.com/watch?v=qOKpWU8vvM8
Hope this help.
Actually, you can use NetLogo simulator for your work. It is a multi-agent-based and can help a lot. I used it for my Ph.D. dissertation and was very useful.
you can design your own server using python flask or others, then use MQTT or other similar M2M communication to receive all the data in your server and apply machine learning algorithm as you wish in your designed server and you can visualize your result on the frontend.
Dear Md Rezaul Kabir Tutul
Thank you for a good idea. looks interesting. However , I have few questions about your idea:
1- I see that flask is used in examples to deploy a trained model as a web app. However, is it possible to train the model on flask also? I need to store thousands of data, train my model on them every night, and then deploy it to predict the next day and show predictions.
2- In this case I am using flask server for my model, right? Is it free to use? as my model and the amount of data are heavy.
3- In case it is not free, can I have a board like NVIDIA to train my model every night and show the predictions online on flask?
Thank you
use SQL or MongoDB database to store your sensor data. then use SVM or knn for your prediction at every 24 hours. flask is free to use. if you think raspberry pi is not able to run your model then you can use google coral tpu (search internet). but I don't think you need this. if you are new I recommend to use raspberry pi. tons of tutorial on the internet.
https://projects.raspberrypi.org/en/projects/python-web-server-with-flask
https://scikit-learn.org/stable/modules/svm.html
- Badges
- Science topic