FilmForecast

Project aiming at Predicting Movie Ratings using Methods of Machine Learning

Topic : Predict the Rating of Movies

Our Approach :

At first, Using a Linear Regression and excluding the unrated movies (Rating == 0) we obtained :

📊 Validation R2: 0.05239199004077433
----------------------
📊 Test R2: 0.05341469876521021
----------------------
📉 Validation MSE: 3.756533272317015
----------------------
📉 Test MSE: 3.7099452036246263
----------------------

With few examples -> only train on a subset of the data, less statistical power When target variance shrinks, R2 naturally drops if the model handle much variation

Using a Linear Regression model including unrated we achieved the following results:

📊 Validation R2: 0.48586217330126347
----------------------
📊 Test R2: 0.4883999483580138
----------------------
📉 Validation MSE: 4.4539062880351
----------------------
📉 Test MSE: 4.418847851586559
----------------------

Implementing a Ridge Regression model we achieved the following results:

📊 Validation R2: 0.48664158818283676
----------------------
📊 Test R2: 0.49493691522476735
----------------------
📉 Validation MSE: 4.4553966021437255
----------------------
📉 Test MSE: 4.370727223787257

! Using a Linear Regression model with Polynomial Features we achieved the following results:

Note : We did not use Polynomials for all features, as it would lead to a very high power consumption. The features we used ['vote_count_log', 'budget_scaled', 'revenue_scaled', 'runtime_scaled', 'popularity_scaled'] `

📊 Validation R2: 0.682862375791018
----------------------
📊 Test R2: 0.6661081283617595
----------------------
📉 Validation MSE: 2.7514282784156125
----------------------
📉 Test MSE: 2.8967269006055836
----------------------

Trying out Lasso Regression, we obtained the results:

- Train MSE: 4.6666
- Validation MSE: 4.6926
- Test MSE: 4.6840