Article: Quality of Data from IOT Devices

IOT devices are said to be creating a revolution in the technology world, not only because of the ability to connect multiple devices which can then perform a useful function or set of functions, but also for their Data Capture ability which then enables Data Analytics, and ultimately leads to improvement of the eco- system that they operate in. This article focuses on the Data Capture part of IOT and more importantly, on what should be the criteria for Data Capture, and the Quality and usefulness of the captured Data.

What can be Captured by an IOT Device

1. Sensor Data

This could be temperature, pressure, voltage, current, sound level, light intensity, etc.

2. Change in State of a system

Toggling of a switch, opening or closing a door, change in level of a container, speed of a car, etc.

3. Number and type of Operations

Different operations and the number of times they have occurred. An employee passing through a door, a change in password, removal of an item at a store etc.

4. Timing Data or Timestamp Data

Time taken for an event from start to completion, time elapsed between events, recording of date and timestamp.

5. User Inputs

Relevant when a Human Machine Interface is part of the system. Page views, navigation, selection of an item, path to selection, time spent on each page or item etc.

Results of Data Analysis

The data captured can give insights on a broader category of things which are more eco-system related, and may help in drawing useful inferences as to what may be a problem in the system, or what could be a path to improvement of the system

1. Environment

So sensor data can give insights about the weather, or power consumption pattern during the day, or noise pollution. An interesting case is the use of mobile phone data to estimate traffic congestion on roads.

2. Events

Change in state reveals the occurrence of events, and this could lead to optimization of the system to react to these events. So an open door could signal that the air conditioning would be inefficient, and so maybe change the blower speed for the duration the door is open.

3. Metrics

Improvement of a process by reducing the number of operations required to complete it. Elimination of inefficient operations.

4. Frequency, duration and Record

A process taking too long or too short, or happening too often, or not often enough may have implications. Say machine oil not replaced for 6 months may cause gear box failure. Compressor in a fridge starting and stopping too often may signal a bad thermostat. Timestamp data of course is useful purely as a matter of record, which may lead to audit trails etc.

5. User Behavior

Analytics is predicting the level of satisfaction users have with an interface, based on how long they spend or the manner in which they navigate through a user interface. This helps design better, user friendly, more intuitive user interfaces.

Quality of Data Sample

There are a few measures by which we can determine if the data collected is any good. Of course a reasonable sample size may be required.

1. Standardization

The data should be in a well defined format, and should correctly represent what was being collected.

2. Repeatability

Data collected under similar conditions should produce similar results. A good example of this not happening is when a hardware dependant app is loaded in two different devices, one with superior hardware than the other. Same environmental conditions produce different data in these devices, because of the difference in hardware.

3. Comparability

Data sets should be comparable, in order to analyze the behavior of eco systems. This means that one input in isolation may not be enough, but only a data set comprising of a minimum number of inputs would give sufficient grounds for comparison.

What should be Captured

To determine what needs to be captured, one must define the role of the IOT device and the context of its use, very well.

Now maybe it has a number of sensors, but if the intent is to measure say the effect of sunlight on the efficiency of an air conditioner in a room, then maybe temperature, light intensity and time of day data are sufficient.

Also data from one device alone may not be enough. There may be hotspots in the room due to change in angle of incoming sunlight, and so multiple devices would need to be distributed across the room.

What can be Left Out

1. Do not measure everything, just because you can

Unnecessary inputs, like inputs from sensors that do not contribute to the task at hand.

2. Avoid recording data too frequently

Study the system to see how often you need to measure inputs. Measuring too often does not give any useful results, and only wastes system resources.

3. Do not record timing data from all devices, working together

This is an unnecessary waste of device memory. Make sure one device records timestamp, and the other devices are synced with this device.

4. Avoid Data Overlap

Too many devices in the same zone, producing similar inputs are a total waste.

5. Unnecessary diagnostic information

Make sure diagnostic inputs are optimized. There may not be any need to record battery voltage every hour. Record only when it reaches within; say 25% of indicative battery life.

While analyzing data also, it is important to focus on the purpose of analysis. It is very tempting to go wild, and draw all kinds of conclusions from the data collected. This is of no use to end customer and only creates confusion.


Too much data is as useless as too little data. Avoid data clutter while deploying IOT devices. Do not let the Noise hide the Message.

Thanks for Reading. Do Checkout Our New Design Packages For Startups

TL Micro Automation - Electronic Products , Embedded Design Services , Consulting