Predicting the popularity of food tweets

Final exam project for the Introduction to Machine Learning course, held by Prof. Eric Medvet (@ericmedvet) at the Univeristy of Trieste during the 2022-2023 academic year.

Problem statement

It is reuqired to study a ML-based technique to predict potential popularity of a tweet that talks about food. The data is not provided, so the part about data collection has to be treated as well.

Proposed solution

After collecting data with Twitter APIs, we learn a dummy regressor, linear regressors, Regression trees and Random forests models and evaluate them in terms of effectiveness (measured throug the Mean Squared Error of the predictions) and efficiency (related to prediction times).

The best model results to be a Random forest with $100$ trees and $50$ maximum observations for each leaf node. At a little cost of effectiveness, even a single Regression tree achieve good results.

A detailed report of the study in can be found in the .pdf file inside the ./relation/ directory. All the R code used in the present project, with the exception of those for the data retention, is available in the .Rmd file.