Bad Sensor Data Is Still Bad Data, Even If You Have AI

Imagine your model receives this:

72.1
72.4
72.6
9500
72.8

There's a good chance 9500 is wrong.

But your AI model doesn't magically know that.

That's why data validation matters so much in IoT.

Sensors can produce:

Missing values
Duplicate values
Impossible values
Incorrect timestamps
Random spikes
Delayed readings

A basic validation layer can catch some of this.

For example:

def valid_temperature(value):
return -50 <= value <= 150

Obviously, the actual range depends on what you're measuring.

And I wouldn't necessarily delete suspicious data either.

Sometimes it's better to mark it:

{
"value": 9500,
"quality": "suspect"
}

That way, you still know something happened.

The more I think about IoT, the more obvious it becomes:

Good AI starts with good data.