Coding.Waterkant2020 - Prediction of ticket submissions C6/7

• 

  Hourly Predictions of ticket submissions with LSTMs

• 

  Daily Predictions of ticket submissions with LSTMs

• 

  Daily Predictions of ticket submissions with ARIMA - long term prediction

Hackathon Waterkant Festival 2020 in Schleswig Holstein

This project was part of the Hackathon of the Waterkant Festival 2020. It was part of a challenge that was created by the Landeshauptstadt Kiel.

Challenge Text

As the Landeshauptstadt Kiel with more than 5,000 employees we try and do our best for all our citizens each and every day. We want everyone to be able to do their best job and we want our internal processes to run smoothly - so we need a strong team in the background who will have our back in case of problems. As a help desk, this team works on up to 1,000 different requirements and problems every day. We need your support so that the help desk team can organize themselves even better.

Our goal was to

• predict the amount of incoming tickets
• predict the lifetime of a ticket.

For the first goal we tried three different approaches, traditional models like ARIMA and Exponential Smoothing as well as modern approaches like LSTM. For the second goal we used a Random Forest Model.

Conclusions

Our goal was to predict the amount of incoming tickets

• The LSTM model seems to be the most promising - for daily as well as for hourly data. Further tweaking should improve the results, as well as additional feature engineering
• However, the traditional approaches for daily predictions are nearly as good as the LSTM

predict the lifetime of a ticket.

The Random Forest Model needs more feature engineering in order to improve the results

Outlook

We like to advance the project together with the LKSH to bring the models into the production system.

Built With

• keras
• python
• r
• tensorflow