Fuel Inventory Reconciliation Modeling for a Large Manufacturer

Case Study

Fuel Inventory Reconciliation Modeling for a Large Manufacturer

Business Objective

Our client is a leading manufacturer of underground fuel tanks. Underground fuel tanks always remain under risks of possible leakage and erratic fuel dispense and delivery. Leakage implies water has seeped from the ground into the tank or fuel has leaked from the tank to the ground.

As part of the regulatory check, the fuel site owner based in Europe had to confirm that all the tanks were performing well. The existing validation process was based on statistical inventory reconciliation, and it raised too many false alarms. This validation had to be performed every week. Too many false positives led the reviewer/analyst to go through all of them and provide recommendations.

The client wanted to build a model that reduces the number of false positives in the current process (Statistical Inventory Reconciliation) of identifying a faulty tank. They also wanted to build a Shiny App to visualize the data and results.


  • Irregular transmission of data from sensors
  • Lack of clear definition of the failed test date
  • Extremely imbalanced sample data (0.3% Target Variable)

Solution Methodology 

  • Analyzed daily data (Daily Loss Gain, Daily Delivery, Daily Stock Level (Opening/ Closing), Daily Temperature, and Daily Sales), alarm data (Sensor Type, sensor state), and stock adjustment data (adjusted amount, adjustment reason) for the past 550 days
  • Identified relevant data using feature engineering and then built, trained, and tested a model
  • Trained the model to negate the calibration errors in the tank and suggest whether it has any leaks
  • Built the model to also factor in the weather data. Built many models, one of which got accredited by the Environmental Protection Agency (EPA)
  • The output of the model is stored in a folder which is picked up by the Shiny app
  • The Shiny app had two personas – Site Master and Analyst
    • Site Master- to upload the data and run the model
    • Analyst – to look at the model output of the leaking tanks and send the confirmation to the Site Master saying that there is a leak or there isn’t

Business Impact

  • The model was able to reduce the total false predictions by 36% compared to the current process
  • The developed Shiny App has come out as a useful tool to analyze the leakage prediction and also to visualize the results
Download Case Study

©2023 Tiger Analytics. All rights reserved.

Log in with your credentials

Forgot your details?