In current electricity market, demand side buys power indirectly through electric utilities, and it makes the demand side inelastic to market's price volatility and significantly affects the efficiency of the market. To fix the problem, it is essential to involve the demand side directly. One major step taken towards this goal is demand response programs. These programs offer many benefits and can provide solutions to many issues in the market. However, some challenges are facing their implementation, chief among them is the estimation of load reduction. An accurate measurement of the load reduction needs an accurate estimate of Customer Baseline Load (CBL). In this dissertation, it is observed that the CBL methods developed for large industrial and commercial customers are not satisfactorily accurate when applied to residential customers. Residential customers have a variable load. Moreover, with increasing penetration of distributed generation and storage devices, large industrial and commercial loads are also becoming variable. Therefore, it is an imperative to explore new methods to estimate the CBL for variable loads accurately. In this dissertation, the challenges associated with the presence of CBL are studied carefully, and a \textit{k}-means clustering method based on the average load and a predictability index is proposed to improve CBL estimation. The advantages of the proposed method are shown both in theory and through an experiment. It is shown this proposed method can improve the error performance of CBL estimation considerably.