### 1. Introduction

### 2. Background

### 3. Previous Work

### 4. Methodology

### 4.1 Data Collection and Analysis

Standby mode: CPU, brightness, 3G, GPS, Wi-Fi, and Bluetooth are the most dependent variables (about 14.4%), with S.E.

_{est}=0.17 with a statistical significant value less than 0.05.Video: Brightness, 3G, Bluetooth, Wi-Fi, CPU, Sync, and GPS are the most dependent variables (about 25.7%), with S.E.

_{est}=0.63 with a statistical value that is significantly less than 0.05.Web browser: 3G, Bluetooth, brightness, Wi-Fi, CPU, Sync, and GPS are the most dependent variables (about 23.3%), with S.E.

_{est}=0.59 with a statistical value that is significantly less than 0.05.

^{rd}order polynomial function. It has been noted that it may be close to the real data sampled. However, using the polynomial order 3 requires three coefficients. To predict the discharge rate, three outputs (for three coefficients,

*x*

^{3}*, x*

^{2}*,*and

*x*) rather than one output (for one coefficient

*x*) are needed.

### 4.2 Creation of the Model for Predicting the Discharge Rate of the Battery Rate

*n*is the total data,

*z*

*is the real value, and*

_{i}*y*

*is the predicted value.*

_{i}*n*is the total data,

*z*

*is the real value, and*

_{i}*y*

*is the predicted value.*

_{i}### 5. Numerical Results

^{rd}order polynomial coefficients. MAE and RSME values were compared for each coefficient. For the 3

^{rd}order polynomial function, there were 3 outputs since there were three coefficients. We are only reporting the average MAE and RSME of all of the coefficients in this case.

### 5.1 MLP Models

Learning rate: This is the value between [0,1]. A lower learning rate will slow down but it may still provide high precision, while a high learning rate may not be good if the data is overly distributed.

Momentum: This is the value between [0,1]. Less distributed data may yield more momentum to provide smooth learning, while more distributed data may require a lower momentum to oscillate around the data set.

The number of hidden layers reflects the linearity of data. A zero hidden layer means linearly separable data.

#### MLP with linear regression

#### MLP with polynomial regression

*x*

*,*

^{3}*x*

*, and*

^{2}*x*), which corresponds to Output 1, Output 2, and Output 3, respectively. Thus, for higher polynomial degrees, the network needs more outputs.

### 5.2 SVM Model

#### SVM with linear regression

#### SVM with the polynomial regression

### 5.3 Comparison Results

### 6. Applications of the Models

### 7. Conclusions

^{rd}order) regression. The two models used were the MLP and SVM. To find effectiveness, we compared using RMSE and MAE for all of the cases. The smaller the value is, then a better model can be used for prediction.