Sample 5: Binary Classification with Web Service: Adult Dataset

January 12, 2016

649 views
58 downloads

Algorithms

Report Abuse
This experiment demonstrates how we can build a binary classification model to predict income levels of adult individuals. The process includes training, testing, evaluating the model and setting up web service on the Adult dataset.
# Binary Classification: Income Level Prediction In this sample experiment we will train a binary classifier on the [Adult](http://archive.ics.uci.edu/ml/datasets/Adult) dataset, to predict whether an individual’s income is greater or less than $50,000. We will show how you can perform basic data processing operations, split the dataset into training and test sets, train the model, score the test dataset, evaluate the predictions, and develop Web Service for the experiment. The complete experiment graph is given below. ![][image1] ## Creating the Training Experiment ## 1. Drag and drop the **`Adult Census Income Binary Classification dataset`** module into your experiment's workspace. 2. Add a **Project Columns** module and use the column selector to exclude `education` because `education` and `education-num` duplicate the adults' education information. So we only keep one feature `education-num` in this experiment. 3. Use two **Clean Missing Data** modules to handle missing values in different features. 1. In the first **Clean Missing Data** module, we select columns `workclass` and `native-country` from column selector, choose _Custom substitution value_ from the Cleaning mode and enter text _Other_ in the Replacement value. ![][image2] 2. In the second **Clean Missing Data** module, we select columns `occupation` from column selector, choose _Custom substitution value_ from the Cleaning mode and enter text _Other-service_ in the Replacement value. ![][image3] 4. Use **Metadata Editor** module to convert the string features to categorical features. In the column selector, we select these string columns: `workclass`, `marital-status`, `occupation`, `relationship`, `race`, `sex` and `native-country`. We also choose _String_ as Data type, _Make categorical_ in Categorical, and remain the rest options as default. ![][image4] 5. Add a **Split** module to create the testing and test sets. Set the _Fraction of rows in the first output dataset_ to 0.7. This means that 70% of the data will be output to the left port and the rest to the right port of this module. We will use the left dataset for training and the right one for testing. 5. Add a **Two-Class Boosted Decision Tree** module to initialize a boosted decision tree classifier. ![][image5] 6. Add a **Train Model** module and connect the classifier (step 5) and the training set (left output port of the **Split** module) to the left and right input ports respectively. This module will perform the training of the classifier. 7. Add a **Score Model** module and connect the trained model and the test set (right port of the **Split** module). This module will make the predictions. You can click on its output port to see the actual predictions and the positive class probabilities. 8. Add an **Evaluate Model** module and connect the scored dataset to the left input port. To see the evaluation results, click on the output port of the **Evaluate Model** module and select _Visualize_. ## Results ## From these results, you can see that the **Two-Class Boosted Decision Tree** is fairly accurate in predicting income for the `Adult Census Income` dataset. ![][image6] ## Set Up Web Service ## Once you are satisfy with the above results, you can set up a web service for this experiment. 1. Select _Update Predictive Experiment_ from the **SET UP WEB SERVICE** option, you will see a _Predictive experiment_ as below. ![][image7] ![][image8] 2. Make some minor modifications on this _Predictive experiment_ before deploying is as a web service. 1. Add a **Project Columns** module that links to the out port of **Metadata Editor** module and the right port of **Score Model** module to remove the label column `income`. 2. Connect **Web service input** to the right port of **Score Model** module so that the format of input data from web service will be same as the the pre-processed data. ![][image9] 3. Deploy web service, you will see a web service dashboard as below. ![][image10] You can also test the results by giving a particular set of input values through the **Test** option on the dashboard. Values of categorical features can be selected from the dropdown menu. ![][image11] <!-- Images --> [image1]:https://raw.githubusercontent.com/mezmicrosoft/Sample_Experiments/master/Sample_5_Binary_Classification_Adult_Data/image1.PNG [image2]:https://raw.githubusercontent.com/mezmicrosoft/Sample_Experiments/master/Sample_5_Binary_Classification_Adult_Data/image2.PNG [image3]:https://raw.githubusercontent.com/mezmicrosoft/Sample_Experiments/master/Sample_5_Binary_Classification_Adult_Data/image3.PNG [image4]:https://raw.githubusercontent.com/mezmicrosoft/Sample_Experiments/master/Sample_5_Binary_Classification_Adult_Data/image4.PNG [image5]:https://raw.githubusercontent.com/mezmicrosoft/Sample_Experiments/master/Sample_5_Binary_Classification_Adult_Data/image5.PNG [image6]:https://raw.githubusercontent.com/mezmicrosoft/Sample_Experiments/master/Sample_5_Binary_Classification_Adult_Data/image6.PNG [image7]:https://raw.githubusercontent.com/mezmicrosoft/Sample_Experiments/master/Sample_5_Binary_Classification_Adult_Data/image7.PNG [image8]:https://raw.githubusercontent.com/mezmicrosoft/Sample_Experiments/master/Sample_5_Binary_Classification_Adult_Data/image8.PNG [image9]:https://raw.githubusercontent.com/mezmicrosoft/Sample_Experiments/master/Sample_5_Binary_Classification_Adult_Data/image9.PNG [image10]:https://raw.githubusercontent.com/mezmicrosoft/Sample_Experiments/master/Sample_5_Binary_Classification_Adult_Data/image10.PNG [image11]:https://raw.githubusercontent.com/mezmicrosoft/Sample_Experiments/master/Sample_5_Binary_Classification_Adult_Data/image11.PNG