Forecasting Techniques for Time Series from Sensor Data

Forecasting has always been of interest. Whether one's field is finance, health or seismology, being able to predict future values based on previously gathered data proves to be invaluable when taking decisions concerning the future. In this paper, we research machine learning techniques for predictions on time series and choose the best models that fit our use case, Smart Farms, in which we distributedly analyze time series received from farm-monitoring sensors. On time series with short term dependencies, like temperature or pressure, we make predictions with Hidden Markov Models, whilst for those with long range dependencies, like ground wind speeds orprecipitations, we use Recurrent Neural Networks with Long Short-Term Memory architecture.