IoT provides big contribution to healthcare for elderly care at home. There are many attacks in IoT healthcare network which may destroy the entire network. A propose a framework may be produced an efficient treatment for elderly care at home with low power consumption. A framework contains three phases names; medical data collection layer, routing and network layer and medical application layer. It intends to increase security performance through prediction and detection attacks in real time. Cooja simulator is used for generating real-time IoT routing datasets including normal and malicious motes based on different types of power. The generated IoT routing dataset using data augmentation (SMOTE) to increase the size of dataset. The preprocessing of the generating dataset using three methods of feature selection which are weight by rule, Chi-Squared and weight by tree importance using random forest reduce noise and over-fitting. A proposed model uses convolution neural network (CNN) to detect and predict IoT routing attacks to identify suspicious network traffic. A number of studies have been carried out in this area, but the issue of the extent of the impact of attacks on energy consumption is an interesting topic. Attacks can affect the network completely, in particular on the power consumption of smart devices. Therefore; the main target of this research is detecting and predicting different types of IoT routing attacks which have impact on power consumption and destroy the entire network. This work analyzes the impact of IoT routing attacks on different power consumption using CNN to achieve low power consumption by detecting different types of routing attacks. The experimental results show CNN can detect different types of attacks that have a bad impact on power consumption. It achieves high accuracy, precision, recall, correlation and low rate in error and logistic loss and this leads to decrease power consumption.