Cloud Prediction using Microsoft Azure Machine Learning

Hi there readers. Today i want to share little exposure about Microsoft Azure Machine Learning which is an web-based data analysis application that run online and do all the processing and data storing on a cloud server. Its using a simplified user friendly GUI and drag-and-drop template which is easy and faster to do certain data processing. For the detail system operation, you need to read by yourself because i'm here about to share on the Data Predicting operation and method. 

1. Account Sign-Up

Microsoft Azure ML is of course an licensed application but it offer a free testing account for about 1 years (not sure depends on the company offer). For starter do sign-up the free account at Microsoft Azure

2.Create New Experiment

For starter, you must create a New Experiment. Click at the NEW button at the bottom left side and select Experiment, then select Blank Experiment. 

3.Create Analysis Model

After choosing the method to be used, we will create the main standard structure for analysis. 

Insert Dataset

Firstly, we will insert a dataset (data to be process). For this tutorial, we will insert a sample dataset first. To upload your own dataset please refer here: Upload dataset from Microsoft Excel. At the left side panel, at the top, click Saved Dataset, then click Samples and select Automobile Price Data (raw) for example. To insert to the work space, drag the dataset into the work space. If you want to know the details and helps on how certain module works, just click at any module in the workspace and click more helps under Quick Help at the bottom right panel.

Select Column

After that, we need to insert few module to process our dataset. To find module easily you can just search at the Search experiment items at top left side panel. Now insert Select Column in Dataset to select the column of data that we will be used for processing. Connect the module with dataset from output to input. After both are connected, we can 

select the column by clicking on Launch column selector at the right Properties panel. 

Then you select By Name, click any data column in the list and press Ctrl+A to select all an then click the right > arrow to move all data to our selection then click Done at bottom right.

Split Data

Next, we will need to split our data to two portion which is one portion for Train Model and other one for Score Model. Click the Split Data module, then go to right Properties panel. Under the Fraction of rows in the first output dataset, set the fraction for the data. As example if we insert 0.7, that means 0.7 will be split on first output (Train Model) and the other 0.3 will be on second output (Score Model).

Train Model

Then, we will insert Train Model first. We will train our data here so that we will achieve significant result at the Score Model. The data here will undergo through algorithm processing that we selected. After trained, then it will next be compared at Score Model. Left output from Split Data will connect to right input of Train Model, and left input is the Analysis Method that we will use. 

After that, click Train Model and go to right Properties panel, click Launch column selector, then select the column that will be trained. For this example, search and select Price, then click Done.

Analysis Method

Before you start creating the analysis model, you should determine which method will you be using and make sure its suitable with your type of data. Every type of method have different type of algorithm that are made for different type of data type. Mainly there is four analysis method that being used for prediction which is:

• Classification
• Regression
• Clustering
• Anomaly Detection

For clarification, you can refer this link: Machine Learning Algorithm

Score Model

After the dataset is trained at Train Model, it will be compared here at Score Model with the original dataset that we split (0.3 fraction). For the left input, connect the output from Train Model and the right input connected to right output of Split Data.

Evaluate Model

Then, at Evaluate Model, the Score Model will be evaluate to form data analysis presentation of the result which is we call Visualize. After connecting all together, you can rename your experiment above the work space and then click Save below. 

4.Run Analysis

After all model and module is connected together, then you can click Run at the bottom panel and wait for few second until all operation is done. Wait until all module is status is Done.

5.Visualize Result

After finish running the process, click Evaluate Model then right click, select Evaluation results, then select Visualize. For this analysis using Regression Method, it will show this type of data and graph. 

For this Regression method, Root Mean Square Error (RMSE) is too high (ideal is 1>X) and some pre-processing to the data need to be improve to get a better results. For Coefficient of Determination (R²)  shows up pretty nice because it should be (1≈X)