This thesis addresses how to further improve the prediction of volatile residential electricity loads in the context of energy communities. New decentralized local community concepts including renewable energy production are central components of the energy transition. An important component of these concepts is accurate load forecasting. The latter enables operating home energy management systems and thus increases local e fficiency and self-consumption. Various cluster approaches exist in the literature to support models by pre-grouping raw input data. In this work, hourly electricity loads for 19 UK households were investigated. Using the k-Means algorithm, repetitive daily profiles were identified across all households. On the one hand, this allowed investigating typical energy consumption patterns at household level. On the other hand, the clustering supported models in predicting more accurately. The effect of clustering daily load profiles on the prediction performance was tested for two time resolutions. Using the Normalized Root Mean Squared Error, Support Vector Regression and the Random Forest Regressor were compared on the different clustered subgroups. The main result of the work is that different daily profiles could be identified across a set of households. With the help of these results, the Normalized Root Mean Squared Error could be reduced. For the prediction at daily resolution, this effect was strongly pronounced, while at hourly resolution of the residential electricity loads, a slight improvement could be achieved.