The quality of finished product of a cement mill is measured in terms of blaine, which is the measure of specific area of cement. Normally blaine is measured offline and maintaining the blaine is very important because it directly hampers the cement strength and also affects production cost. A soft sensor based kernel autoregressive exogenous model (ARX) was developed to predict the blaine quality for a defined sampling period to be used in a controller. ARX model includes the past blaine predictions as regressors in addition to the other informative variables in order to predict the blaine. The quality of predictions is largely dependent on data; the construction of data to be used in the algorithm requires good process understanding as the raw data collected from the process will have many information that can mislead the prediction. This means the information may cause over fitting or sometimes reverse modeling because of excess information. In this paper, an automatic method to align data based on the process characteristics to be fed into the algorithm for improving the prediction based on data reconciliation method is proposed. Data Reconciliation (DR) is a technology that uses process information (input data's) and mathematical model to automatically align the variables according to the dynamics of the industrial processes. © 2006-2016 Asian Research Publishing Network (ARPN).