Prediction of Domestic Water Leakage Based on Consumer Water Consumption Data

摘要

Detection of domestic water leakage normally involves technical investigation and currently is less likely to be conducted remotely. This paper proposes a novel approach to utilize consumer's water consumption data to predict possibilities of water leakage occurrence at residential premises. Simulation datasets containing attributes related to consumer's water usage were created based on primary data collected from consumers. The generated simulation datasets were analyzed using selected classification methods available in WEKA Machine Learning Software. The result showed that 16 out of the 33 selected methods produced classification accuracy greater than 90 percent. Analysis conducted on the best eight methods showed that most of the methods produced very consistent results. Further analysis based on standard error rate also showed similar outcomes. The findings clearly suggest that consumer's water consumption data has the greatest potential to be used for remote prediction of domestic water leakage cases.