Modern smart homes would be equipped with ZigBee sensors that connect home appliances via IoT network. Forecasting the future use of energy for the home appliances would be useful and practical for the home users. Since IoT sensors are designed to collect information in real-time from the home appliances, that include energy usage, indoor/outdoor temperatures and relative humidity measures, the data for harvesting insights should be abundant. Computationally a challenge is to seek for a most appropriate time-series forecasting algorithm that can produce the most accurate results. The difference between the traditional time-series forecasting algorithms and the one that involves IoT data is the ability to learn from the sheer volume of IoT data, which is known as big data nowadays. The sensor data can amount to a huge volume, and the energy drawn from an appliance, for example, air-conditioner can depend on multiple factors – the temperature/humidity of surrounding regions as well as the current weather at the time of the day. In this paper, such forecasting is tested with a range of time-series algorithms including the classical ones in comparison with deep learning which is acclaimed as a suitable prediction tool for learning over very non-linear and complex patterns.