PREDICTING FUEL FLOW IN AIRPLANES

Fuel is of utmost importance as it is one of the non renewable resource and using it in a efficient and smart way is highly required in today's date. This is the most common problem faced by the airline industry these days .Fuel constitutes around 30% of the operating cost of airlines due to which we have higher cost of tickets . Developing cost saving strategies especially on fuel is of prime importance to airlines and reducing the emission of staggering amounts of greenhouse gases along with reducing fuel intake can have a significant positive impact on the environment .Given an idea of the utility of the resources used by my work so that no extra resources are used and may go in vain Now the problem statement is that The ability to predict the Fuel Flow (FF) rate of airplanes during different phases of a flight (Taxi, Takeoff, Climb, Cruise, Approach, and Rollout) will help understand ❖ The significant drivers of FF rate for each of these phases and also help understand the factors that make the airplanes perform at higher levels of fuel efficiency during the different phases of a flight. ❖ Insights from the exercise can help derive the best practices, which make flights more fuel efficient under different conditions. After the project it was observed that climb, cruise and approach are most important phases for optimizing fuel and the consumption. The most important features consisted of rate of change of altitude, longitudinal acceleration and ground speed. The clearest visible trend between predictors and fuel flow rate is in the climb phase. In other phases, some of the predictors have a weakly visible trend, but since the root mean squared error is small, it is assumed that the features have strong nonlinear interactions which are not clearly visible in simple plots. All of the work was done in Jupyter Notebook using Python and with the concepts of supervised Machine Learning with various data cleaning techniques from tidying up the data to applying different models like Random Forests and eXtreme Gradient Boosting to get the best suitable results and then validation and checking performance using Root Mean Square error (rmse) method.