You can use Machine Learning algorithms to predict sensor data. Simply, take some real values from sensor and then apply regression technique. You can apply advanced ML Algorithm if required then.

May consider analyze the data using Jupyter notebooks with Spark SQL and Matplotlib.

It depends on your IoT sensor data. If acquiring data is time series you can use deep learning (area of Machine learning) methods such as MLP(Multi layer perceptron), CNN (Convolutional Neural Networks) to predict your data.

I did used prediction not in an IoT system, it was in a HET-NET system, i have used Markov model order 2. It provides good results, try to see if you can adapt it in the IoT systems.