There is an exponential increase in the demand of the energy due to increasing electrical devices. This results in an increasing demand versus supply gap. Due to scarcity of fossil fuels (e.g., oil, gas and coal), global environmental concerns, the rise in demand and addition of multiple efficient power generating systems; reformation of the current energy system is imminent. Smart Grid (SG) is introduced to handle above mentioned challenges. Moreover, for the efficient use of SG, exact prediction about the future coming load is of great importance to the utility. It helps the utility to produce as much energy as needed. The objective of this work is to handle the load need in an adequate manner through coordination among appliances in a Smart Home (SH); and real-time information exchange between user and utility. In this research, we proposed two new home energy management systems that are using load shifting technique for demand side management to improve the energy consumption pattern in a SH. This work assesses the behavior of advising plans for real-time pricing and critical peak pricing schemes. Two different models for the scheduling of home appliances are proposed in this research. Both the models focuses on hourly scheduling of appliances in a SH while aiming daily electricity cost reduction, Peak to Average Ratio (PAR) minimization and user comfort maximization. Both these models are implemented at the electricity management controller level, installed in a SH within a SG architecture. In the first model the proposed scheme performance is compared with the crow search algorithm and Jaya algorithm. In the second model proposed scheme performance is compared with the strawberry algorithm and the earthworm optimization algorithm. The proposed schemes performance is assessed for PAR, user comfort and cost. Furthermore, we worked on forecasting load demand at the utility end, for exact required power generation. We used Extreme Gradient Boosting (XGBoost) for load prediction for the next 30 minutes using previous 7 days data recorded at the rate of 30 minutes time lag. For forecasting, in first step we use XGBoost for calculating feature importance, which is then used for feature selection. In next step we use XGBoost for forecasting the electricity load for single time lag, using the selected features. XGBoost perform extremely well for time series prediction with efficient computing time and memory resources usage. XGBoost based load prediction model performed very good for mean average percentage error metric.