Time series forecasting plays an important role in many day to day applications, and is often used as a tool for planning in many areas. In this paper, we propose a generic methodology for time series forecasting. We use a subset of the dataset to build up the system model by compressing the information through clustering and coming up with inherent patterns in the data. These patterns are represented as curves that any timeseries from the given set is expected to follow. It then facilitates the forecasting through linear regression by matching to the closest pattern to each time series that has to be predicted. We applied this approach on Kddcup 2003 dataset for predicting the citations of the research papers and found the results to be on par with best results.