implementAI-CAEchallenge

The Goal The dataset contains of a maneuver being performed by various pilots. Under this maneuver, 10 features like velocity, altitudes etc are measured over time creating a time series for each. Each pilot is then tagged for being competent or incompetent. We were to find patterns in this multi-dimensional dataset that revealed soft skills dependence. This was a challenge, since I had never worked on such a dataset before. It was also amazing because this is the kind of dataset that can exist in any industry and therefore the skills I learn will be much more widely applicable.

What I Did While no concrete model could be built, many approaches to look at the data were tried. A lot of modelling methods attempted like simple linear regression, Gaussian Process Classifier, RandomForest, PCA, Kmeans etc. All detailed attempts in Jupyter Notebook at the GitHub page.

Challenges I faced This entire exercise was new to me. So was definitely a challenge to think about such a dataset. The number of methods applicable could be numerous, but each have their own caveat which needs to be taken care of while implementing. Understanding uncertainities was definitely one big challenge. More importantly, I focused on finding THE method, instead of going deep into one method and applying it consistently. So a lot of time was wasted in jumping from one method to the next, sometimes not even completely coding up the previous one.

What I Learned When I first looked at the challenge, my immediate response was to look at one straight method that could answer the question, but ofcourse I quickly learned that I cannot aspire for those in real life problems. Then even with the methods at my disposal, I could not estimate error bars on the data which I think would be important for determining trends. I learned that I was actually having fun doing this entire exercise and that I should explore more of this!

I am glad this will be a good starting point of my journey. I am proud to have even attempted them :D

Built With

• jupyter-notebook
• python
• sckitlearn