Dependency Tree-Integrator Method for Reducing Mislaid Data Errors in Water Demand Prediction

The water supply demand has grown significantly high par with the increase in population, activities, manufacturing, etc. in recent years. A wide range of research on water demand forecasts based on the accumulated statistical data is formulated to ensure sufficient recommendations on required water supply. Considering these facts, the problem of mislaid water data statistics impacting prediction accuracy is addressed here. This article proposes a Dependency Tree-Integrator Method (DTIM) using linear XGBoost learning to address the aforementioned issue. The proposed method generates two linear dependency trees based on the increasing order of water utilization and equivalent demand from the daily data collected. The presence of mislaid data breaks the linearity of the dependency tree, where classification is performed. This classification differentiates the demand and utilization to regularize the demand. The regularized inputs are integrated to forecast the difference in previous and current utilizations as demands. The dependency tree and the classification processes form the integrity of the linear XGBoost learning. The dependency tree fails if the linearity is interrupted by the mislaid data input that is identified as an error and the differentiation occurs. Through real-time data and statistical analysis, the proposed method’s efficacy is verified using accuracy, precision, mean error, and complexity metrics. Thus proposed DTI method improves prediction precision and accuracy by 11.23% and 11.87% under the different quarters considered.