To work on a task in a new field that we had not worked on before. Additionally investigating how well this challenge can be solved and if it can bring a benefit to surfers.

What it does

The model we developed tries to predict the pageviews from the data it has been given. <\br> The pageviews were the metric we used to inidcate a good surfing day.

How we built it

We loaded and preprocessed the data. This includes:

  1. Combining different years of data
  2. Converting the data to usable standards
  3. Combining and dropping features (for example combining wind speed and direction to a wind vector)
  4. Creatning categorical variable for month and weekdays <\br> After that we created a Feed Forward Network using Tensorflow and trained the data. For concrete information the given github repository can be checked.

Challenges we ran into

The pageviews were not very constant over the years and did also not correlate very well with the weather data. Therefore we had some noise from the weather changing over years and some other factors that negatively influenced the model prediction. <\br> Another challenge was to merge and investigate the data in this short time and then also develop a good model.

Accomplishments that we're proud of

We created a good suited model that could make good predictions regarding the available data. <\br> We learned very much from this project and had good teamwork.

What we learned

Each of us got a better understanding of data investigation, feature engineering and model building for neural networks. <\br> We also learned that the most important part of this process is the pre processing of the data and that this part needs time.

What's next for Prediction of Surfing Conditions

As an outlook for the project it could be enhanced by changing the metric of good surfing conditions. If something could be found that is more suited than pageviews, a model could make much better predictions. <\br> In this case our data preprocessing could easily be used and only the model needs to be newly evaluated and modified.

Built With

Share this project: