Prediction of Consumption of Petroleum Products

India is the world's fourth-largest petroleum consumer. There was a strong volume growth in Petroleum Consumption of India which is now slowed down in the recent two years. This will soon will be reflected in global oil consumption growth. According to "The Economics Times of India"- Domestic consumption data released by the Petroleum Planning and Analysis Cell shows the growth in consumption of petroleum products, which was 5% in FY12 and 4.9% in FY13, slumped to 1.6% in the April-June 2013 quarter. The data shows that only decontrolled products such as petrol, aviation fuel contributed to volume growth. “Excluding minor decontrolled products (Petcoke & others representing 11.1% of total in quantity terms), which are insignificant in value terms, the growth in consumption fell 3.1%," according to the analysis cell.

There are number of factors which are responsible for this slow down including some economic reasons. By predicting the petroleum product usage we can predict the habit of usage of Indian people and hence help in raising the Indian Economy. Azure Machine Learning can help us in this.

To solve this real world problem I have used the data which is made available by the Indian Government for research and analysis on their site.

This dataset has a huge data with parameters such as Light Distillates – LPG, Light Distillates – Petrol, Light Distillates – Naphtha, Middle Distillates – Kerosene that helps us to evaluate water quality.

I have used Machine Learning in Azure and processed this data that will help us to predict the Indian habit of using the petroleum products with quantity.

Technical Architecture

In Azure I have selected Data Analytics and Machine Learning. Then created a ML workspace. Then in ML studio I created a new experiment. The technical architecture is:

1. Uploaded data.
2. Build and validate a model.
3. Created a web service that uses your trained models to make fast, live predictions.

Figure 1: Live predictions

Solution Details

After creating new experiment in ML Studio:

1. I have uploaded the dataset from https://data.gov.in/.
2. Then I begin by identifying columns that add little-to-no value for predictive modeling.
3. I define values which are non-continuous by casting them as categorical.
4. Cleaned data, we must make sure our dataset contains no missing, “null”, or “NA” values.
5. Model Building.
6. Training the Model.
7. Model Evaluation.
8. Published to gallery.
9. Set up the web service.

The prediction can help to predict the Indian habit of using the petroleum products with quantity.

Relevant screenshots of services used from the Azure portal


Figure 2: Update a New dataset


Figure 3: Evaluate in Machine Learning


Figure 4: Evaluate Model


Figure 5: Score Dataset


Figure 6: Save trained model


Figure 7: Dashboard


Figure 8: Enter Data to Project


Figure 9: Summary and Description

Azure Machine learning helps up to solve the real world problem. With Predictive Analysis we can predict or recommend solutions. We can also publish this model as Web Services and to Azure ML Gallery.