Afshin Afshari

Handling class imbalance in customer behavior prediction

Class imbalance is a common problem in real world applications and it affects significantly the prediction accuracy. In this study, investigation on better handling class imbalance problem in customer behavior prediction is performed. Using a more appropriate evaluation metric (AUC), we investigated the increase of performance for under-sampling and two machine learning algorithms (weight Random Forests and RUSBoost) against a benchmark case of just using Random Forests. Results show that under-sampling is the most effective way to deal with class imbalance. RUSBoost, as a specific algorithm designed to deal with class imbalance problem, is also effective but not as good as under-sampling. Weighted Random Forests, as a cost-sensitive learner, only improves the performance of appetency classification problem out of three classification problems.

MID-TERM FORECASTING MODEL OF ABU-DHABI'S ELECTRICITY CONSUMPTION APPLIED TO DEMAND-SIDE MANAGEMENT IMPACT ASSESSMENT

Climate change, pollution, reduced infrastructure investment availability and escalating fossil fuel prices have resulted in renewed emphasis on energy conservation and efficient electricity infrastructure utilization through Demand Side Management (DSM) in the existing building stock. DSM measures ranging from enhanced building controls to equipment/envelope retrofits are designed to address this problem. The difficulty to accurately assess the ex-post impact of such measures is a widely recognized barrier to the wider deployment of DSM. The task is complicated by the dynamic nature of the energy consuming processes, the coupled interaction of multiple subsystems and the high correlation of demand with weather and other perturbations. An hourly regression-based model of the load, driven by exogenous variables is proposed to address this problem. The model was estimated for the city of Abu Dhabi, UAE, using measured data from preDSM period. It was then used to profile the “baseline” energy consumption over a selected post-DSM period revealing, though comparison with the actual energy consumption, the savings attributable to the DSM intervention. The model produced accurate results; adjusted Rsquared of 0.9931 (training period - year 2010), a RMSE equivalent to 1.84% of the annual peak load, and a MAPE of 2.64% (verification data-set first-half 2011).