Finding unexpected value from a complex set of data is one of the benefits of data science, although you will generally find it more useful to spread some intelligent decision making across a large scale or at a high velocity. But sometimes, you will find some unexpected and creative value from seemingly mundane data. A wild example is depicted in the following image, which was generated by a Google deep learning algorithm:

After you understand what the IoT devices are reporting, combine it with external data, identify interesting patterns, and communicate the information to others, you can add additional value with geospatial analytics.

IoT devices are typically geographically distributed and can even sometimes move around. Incorporating that positional and usage information can add new opportunities for analytics and valuable services. This can also add additional features that have predictive value. Methods to leverage these techniques were covered in Chapter 9, Applying Geospatial Analytics to IoT Data.

Data science is a broad field with some exciting recent developments in the application of machine learning techniques. The key concepts of machine learning were covered in Chapter 10, Data Science for IoT Analytics, along with a couple of popular algorithms: random forest and gradient boosting machines.

Feature extraction, the bias - variance trade-off, and validation were discussed. Trend forecasting using ARIMA with R example code was also introduced. Deep learning was discussed as a potential method to wring extra value out of IoT data, if you have the right data.

In short, we covered the following recommendations:

• Use location to learn useful things from your IoT devices
• Build a rich set of features for machine learning
• Be careful when applying machine learning techniques but definitely explore their use
• Use deep learning when it makes sense and don't use it when it doesn't