Demand response is a reduction in electricity consumption designed to prevent system emergencies stemming from demand spikes during peak periods. While demand response has been embraced as a cheaper alternative than adding peaking generation, the implementation has been challenging, especially for residential customers. The peak period baseline load of residential customers is difficult to estimate due to the load pattern randomness emanating from weather or behavioral variations. In this thesis, a novel clustering-based customer baseline load (CBL) is proposed to improve the error performance of the traditional baseline estimation methods. The proposed method assumes there are true underlying clusters of consumption profiles of residential customers, which differ only with respect to some feature(s). The spectral features obtained, via Shannon entropy (SE) estimates, from maximal overlap discrete wavelet packet transform (MODWPT) decomposition of the historical consumption, were harnessed to compute a set of new CBLs for existing baseline methods. The proposed method shows significant error performance improvement with respect to peak period baselines. The thesis is extended to a case study of a Dynamic Peak Rebate (DPR) pricing demand response program. The amount of rebate payment was estimated by clustered linear regression (CLR). Finally, the demand reduction costs of the DR event load reduction are calculated for various CBL estimation methods. The proposed CBL method comparatively provides the lowest demand reduction costs in all the DR events considered.