Fault Prediction in Wireless Sensor Networks using Soft Computing

With a deluge of data from various sources, sensors being resource constrained with less computing power, small memory and battery life leads them into getting compromised with various attacks and malfunction. Failures or Fault tolerance based on a single source of information has inaccuracies. Faults should be identified and described by multiple feature of data collected from different sources over a period of time. Generally, fault tolerance has been carried out by monitoring the health of each device which in itself consumes a lot of battery power. Also we need to know what should be an apt time interval or frequency to consider predictive analysis. Predicting faults in wireless sensor networks is a challenging proposition that requires deep study of the past data available. With advances in Machine learning and data analytics in particular have changed the way organisations can approach this issue to monitor their health and take preventive measures. In this paper, we propose a soft computing-based sensor node fault prediction based on time-series data. Using ARIMA, we predict failures by decomposing time series data into trend, seasonality and residual components and fit a model. A robust forecast estimation model with confidence bounds: 80% and 95% is developed which can help in implementing an intelligent IoT infrastructure in WSN. There is more ambiguity regarding the longer term forecast as the model is supposed to regress future values based on previously recorded predicted values.