Barclays_Airfoil_Noise_Problem

Inspiration

I have been learning data science and data analytics for some time. This hackathon provided me a valuable opportunity to attempt a project on my own and consolidate all that I have learnt so far.

What it does

This project is based on the dataset from the UCI Machine Learning Repository, that looks into building a model that can accurately classify an airfoil based on its output noise level into one of 3 classes : low sound level, moderate sound level and high sound level. Several machine learning models are attempted and the results compared to determine the model with the highest accuracy.

How we built it

Data was loaded into jupyter notebook and using pandas and sklearn packages, data exploration, transformation and model training was performed. Results were evaluated and the best model was found to be decision tree classifier and k-nearest neighbours classifier.

Challenges we ran into

• As I was new to MS Azure, I had to spend time learning about the functionalities available. However, I was not able to learn about the required functionalities within the competition time period and adopted to complete the project using Jupyter Notebook.
• Time management was a challenge to complete this project whilst managing several urgent projects at work.

Accomplishments that we're proud of

• I managed to complete the requirements of the hackathon and present the results of my findings.

What we learned

• It takes time to analyse the data and the direction / problem may not be apparent at first instance. Multiple iterations are required to better understand the dataset and how to interpret the findings.

What's next for Barclays_Airfoil_Noise_Problem

• Explore ways to identify airfoil characteristics based on lower sound output. Even lowest observation is still above acceptable human tolerance (which is typically 80 decibels).
• Project can be further extended to create a UI for users to input parameters of a specific airfoil so as to predict the class of sound level.
• Possibility to use Microsoft Azure solutions for storage of data and perform model training if data set becomes larger.

Built With

• python