Predicting Video Game Hits for 2016
Taking a data set containing information about video games prior to 2016 and training a machine learning model to predict which games have the best chance at becoming hits (sales greater than 1,000,000) while simultaneously trying to figure out the top ten determinants to what may make a video game a hit.
Link to GitHub
Some Insights
Cross-platform makes a difference. More consoles = more potential customers
Critics (who usually get an advanced copy) determine what will do well… or maybe just know their stuff.
Most Important features (top 10):
Critic_Score (0.180300)
PC (0.085355)
Num_Platforms (0.083465)
is_franchise (0.071425)
PS (0.064140)
Publisher_Eidos Interactive (0.036991)
Year (0.027567)
PSP (0.017810)
Franchise_kingdom hearts (0.014210)
Franchise_lego (0.013884)
How did we do?
| Name | Hit Probability | Hit | |
|---|---|---|---|
| Battlefield 1 | 0.973 | Yes | |
| Lego Star Wars: The Force Awakens | 0.970 | Yes | |
| Call of Duty: Infinite Warfare | 0.960 | Yes | |
| LEGO Marvel's Avengers | 0.959 | Yes | |
| Forza Horizon 3 | 0.947 | Yes | |
| Titanfall 2 | 0.947 | Yes | |
| Tom Clancy's The Division | 0.946 | Yes | |
| Overwatch | 0.923 | Yes | |
| Assassin's Creed The Ezio Collection | 0.916 | No | |
| Just Dance 2017 | 0.900 | No | |
*Out of our top 10 most likely games we predicted 8 correctly
Teammate credit: Alex Mical • John Caudle