Model Predictions

Utlizing the data available, it is possible to formulate an accurate model for prediction of sales. By reshaping, encoding and grouping variables into classes, even more accurate perdictions can be made.

The accuracy score tells how many predicted outcomes the model got right. It appears the models are picking up on something.

Data includes all critic and user scores, with weighted score calculations.

Feature Enhancement

The new categorical features are encoded and added with the score features to improve accuracy.

It appears the multi-layer neural network method is putting out solid outcomes. At 87% accuracy, it is our most reliable model. The output is given as either a potential Top Seller or not, which is defined through an algorithm that parses global sales records.

Through use of information such as genre, platform, year of release, and numbers for critic and gamer scores we have a model that can accurately predicted if any particular game will be a smash hit.

After tuning the inputs and layers of the neural network, the model was able to come up with 89% accuracy in predicting video game sales. And with a mean absolute error of 0.15, it is our best model.