Sensor network driven novel hybrid model based on feature selection and SVR to predict indoor temperature for energy consumption optimisation in smart buildings

Energy is a vital resource for smart cities and smart buildings. Maintaining Indoor temperature to a comfortable level requires energy consumption which comes from several resources renewable or nonrenewable. The indoor temperature depends on several parameters such as environmental attributes, building design and structural attributes, user activity etc. Predictions on energy consumption and indoor temperature have employed few data-driven model and machine learning approaches which require more accuracy, better generalisation and real-time estimation of energy and the indoor temperature. This paper proposed a hybrid model based on feature selection methods such as feature importance and support vector regression (SVR) to predict indoor temperature in a more accurate and efficient manner. The best model SVR gives MAE equals to 0.0174, MSE equals to 0.0006, RMSE equals to 0.0256 which are the least error in the prediction of indoor temperature. Random forest regressor (RFR) model gives MAE equals 0.0820, MSE equals 0.0101, RMSE equals 0.1006. RFR results are the worst in all prediction error matrices. Accuracy and efficiency have been compared with other machine learning models and the proposed model has outperformed them.