Predictive Stream Analytics for Threshold based Approach:A Case Study of Temperature Anomaly

An important aspect of daily life is sensing environmental factors within a smart home environment, the majority of which humans use for thermal comfort. A good prediction of these parameters may help the smart home occupants keep things in check and take appropriate decisions in real time. In smart homes, temperature monitoring is a vital parameter so that each occupant may feel comfortable during their stay at home, and since people require different values for the same variable, there may be chances for an anomalous temperature trend, which can lead to discomfort for the occupants. In this paper, we propose an algorithmic approach for unusual temperature trend detection and prediction from streaming data based on a threshold parameter. The data is processed through the Apache Kafka producer and consumer terminology before being passed to flink library for further rule processing and prediction for the next seven days based on the prophetic forecasting model. If the algorithm detects three events with unusual trends within a window of a week, then the user is alerted. The entire workflow is passed through an event processing mechanism to calculate the dynamic threshold value and make an impactful decision. The algorithm is evaluated using a statistical evaluation metric that has an error rate of (i) Mean Absolute Percentage Error(MAPE):0.38(ii)Mean Absolute Error(MAE):1.15(iii) Mean Squared Error:1.34.