Movie Prediction Model (Kylan, Reed, Sean)

Inspiration: To apply the skills we learned at ML@UVA to a real-life project. We are currently developing a LSTM to predict real estate value by analyzing weather patterns and took on this challenge to test the technical knowledge acquired from our club.

What it does: Our code selects the 100 most important features (through selectKbest) and runs a neural network through these parameters to predict earnings.

How we built it. Our first step was to look at what literature was out there. We started by researching important features related to movies that have the highest impact on gross. Most of the features were already in the given dataset from kickoff, however, “movie budget” – which had the highest correlation to high movie grosses – was not provided. To add the budgets of every movie from our dataset, we used an API from the following source, https://developer.themovied

b.org/docs/getting-started. Afterward, we searched existing models for ideas. We came to the publishing, Movie Revenue Prediction Using Machine Learning Models, https://arxiv.org/html/2 405.11651v1, and looked at some of their algorithms. Ultimately, we decided to pursue a traditional neural network.

Challenges we ran into: One of the main challenges was trying to incorporate a gradient-boosting model for this project while still using a torch. Torch didn’t have a gradient boosting model and developing one from scratch was not an efficient use of time during this tight deadline.

Accomplishments that we're proud of is mainly developing a functioning model as beginners in coding. All steps throughout model development were difficult for us as beginners and creating a finished product was a great feeling.

What we learned: This was one of our first models and set the foundation for setting up and beginning future projects. Furthermore, we learned about statistical approaches and K-means for analyzing what parameters contributed most to our prediction.

What's next for Movie Prediction Model (Kylan, Reed, Sean): In the future, we plan on developing a gradient boosting model from scratch that can be used adjacent to torch.

Built With

• pycharm
• python
• selectkbest
• tmdb-api
• torch