Data Cleansing During Data Collection from Wireless Sensor Networks

Quality of data in Wireless Sensor Networks (WSNs) is one of the major concerns for many applications. The data quality may drop due to various reasons including the existence of missing values and incorrect values (also known as noisy or corrupt values) that can be caused by factors such as interference and machine malfunctioning. A drop in data quality may seriously impact the performance of decision support systems. Thus, it is crucial to clean the data before using them. In this paper we analyze the impact of missing values in a WSN data set (which is collected using a Voronoi diagram based network architecture) for the data mining tasks such as classification and knowledge discovery. While the quality of the data mining output (classification accuracy) suffers from the existence of the missing values this study shows an improvement when the missing values are imputed through our data cleansing scheme. The proposed scheme uses a corrupt data detection technique and a missing value imputation method for cleaning the data being collected from the sensor nodes. Our empirical analysis indicates the effectiveness of the proposed approach.