Predictive analytics is a method to make prediction for an unknown event. In the context of IoT, we can develop predictive analytics to make a decision-based streaming sensor data. This is a part of machine learning study. In general, we can make predictive analytics using a diagram that is shown as follows:

Defining business problems is the first step to develop predictive analytics. Some problems probably need experts to make clear those problems. For instance, economics, biology, and volcanology. 

We also should prepare data in order to develop a model. This data should have high impact factors on the model.  When we develop a model, we also perform some steps such as defining targets, extracting derived features from data, fitting the model, and evaluating the model. In a real project, we probably make some iterations to ensure the corrected model.

After we developed the model, we can deploy the model into our system. This could be deployed in web application, RESTful app, or an other platform. We also monitor how the model runs. If we find some errors in the implementation, we can modify our model and data. Then, we perform normal iteration for developing a model.

To implement a machine learning model and then evaluate the model, we can use Amazon Machine Learning. You can access it using your existing AWS account on https://console.aws.amazon.com/machinelearning.

Next, we will develop predictive analytics for an IoT project case using Amazon Machine Learning.