Recession Prediction

February 14, 2015
This experiment demonstrates on how to use AzureML to predict recession using leading economic indicators. 
This experiment uses CPI, trade balance, GDP, and DOW data from *Bureau of Labor Statistics and US census* to predict if the current economic indicators indicate a recession or not. This experiment is designed to only illustrate Azure ML toolset using a small dataset as a sample and by no means provide a real or true prediction. This experiment also demonstrates how to manually enter data, use project columns for data selection, modify data type, merge columns, and normalize data. Finally, it demonstrates how to train the model and use the trained model for predictions ## Dataset For this experiment, we crafted a dataset from the data available on *[Bureau of Labor Statistics]( and [US census](* web sites. This experiment gathered CPI, Trade balance, DOW, and GDP data from these web sites. Once data was available, we used *project columns* to discard the unused data columns, *meta-data editor* to override default data types, *normalize module* to adjust the numeric data on the same scale factor and finally *split* data into training and test datasets. ## Model We use a 80-20 random train/test split for the experiment. We tried using multiple **two-class classifiers** but the best results were observed from **two-class decision forest** algorithm. ## Experiment Details ![Training Experiment Graph]( ## Results ![Model Evaluation]( ## Additional Resources Once the training was completed, we created a prediction graph to do the real time predictions. We also published a web service for integrating with app or excel for real time predictions. [Here is the link to the prediction graph]( Created by a Microsoft Employee