Inspiration
Steam Games are an important part of our lives as gamers. We want the best games to be available in the market. How can we ensure developers design games we as players want? A new model to how many people will buy a Steam game is a great way to show developers if the games they think about would be liked or not.
What it does
The model classifies games (specifically with all the information of tags, playtime, achievements ... ) into either range of expected owners = 0-20k, 20k - 200k, ... or will attempt to fit the number of owner, which is estimated on a previous analysis of the number of owners per review.
How we built it
We performed data cleaning on two raw steam game datasets and merge them together. We tried two approaches. The first sets the number of player (the value we wish to predict) into 5 categories. The second estimate the numerical number of player using a formula in previous studies. Different variable selection approach is performed on each dataset. The categorical data uses random forest to select the 30 variables that matters, while the numerical data uses Lasso regression to select a subset of predictor. Then, for each set, we tried random forest, regression, and neural network to model the number of player that will buy the game base on the length, price, theme, and other characteristics of the game.
Challenges we ran into
The model is not very accurate no matter which approach we choose. The thing is, two game with same characteristics might ended up have very different sales. There are a lot of variables that can not be quantify, which lead to the large random error in the dataset.
Accomplishments that we're proud of
We really learned a lot in the process and we are proud for finishing this project.
What we learned
We learned various modeling techniques and data cleaning procedure that is extremely helpful for data processing.
What's next for Will Players Love Your Steam Game?
We will try to incorporate more data into our model. The current dataset does not have enough variables to completely describe a game.
Log in or sign up for Devpost to join the conversation.