Microsoft Azure has an amazing tool which can enable Business Decision makers to connect the dots and use data associated with Key Metrics (discrete / continuous / Boolean) connect to the business variables which have an impact and build models for predictive analytics.
This tool is the Classic AzureML studio ( link: https://studio.azureml.net/ ). In order to build definitive models using this, you as a decision maker need to master some few select steps to start. This in no way will make you a Machine learning expert, but will help you connect the dots, and help you guide your team in the best possible direction towards building valuable ML models for your decision making. The steps are as follows.
- Identify type of Business Decision and connected business parameters.
- Understand the Business decision metric – Boolean or Discrete. Example: Do you wish to predict store sales or a possible of product outage on a particular day in a store. The other business need would be to categorize your observations into different clusters for business evaluation.
- Identify possible business parameters connected to the metric: Identify all the possible business parameters connected to the metric above.
- Identify data stores: Identify data stores where the metric and associated business parameters are stored. Get ready to collect a lot of data for the same to experiment on ML.
- Data Collection & Readiness:
- Identify your data sources and import mechanism.
- Be ready to import the data.
- Cleansing (removal of duplicates, elimination of missing values etc.)
- Training data Vs Testing data split:
- You will need to train your machine learning model for it to start intelligently identifying patters and building internal algorithms for prediction. For this you will need to split your data into training data (typically 80% of overall data) and testing data (remaining 20%).
- Stratification for Splits: When splitting data, you can use certain business parameters which we can call stratification parameters to split data uniformly. Example if you have data by cities you can use city as a stratification parameter to ensure 80% of observations of each city are used to train the ML model.
- Experimenting with the Models:
- Microsoft has provided a nice cheat sheet for AIML enthusiasts and decision makers to have a starting point when experimenting with ideas for ML models.
5.Using the above sheet, and diving into details of Models for
- Classifications (two class Vs multi class).
you can then select the right model to experiment with. As a simple starting point for discrete/continuous KPI’s regression/clustering will be the direction you might be going in, and for a Yes / No prediction Two class classification is where you will go.
6.Scoring the Results and Evaluating the Model: After your model is trained, the AzureML studio will allow you to use your testing data (the remaining 20%) to see how well your model performs using scoring and evaluation steps.
- Scoring : Allows reviewing predicted values Vs actual
- Evaluation: Uses AIML based metrics to actually give your model a performance score.
AzureML studio allows you to do all of the above steps very easily. See Sample steps actually implemented below.
Try it. This will help you connect the dots and make a lot of valuable business decisions.