1 of 21

VIDEO GAMES SALES MODELLING

Darius Ramli – Data Science Batch 17

6.53

2 of 21

ABOUT ME

My name is Darius and currently I am full-time employee of one of big-four accounting firms. I specialized in Transfer Pricing issue for both local and multi-national corporation.

My aim in joining Dibimbing Bootcamp is to keep my abilities on track with the current data-driven business.

I hope this presentation could show how well I grasp the big picture of data driven business for my future endevour.

Video Games Sales modelling

2

24/03/2023

3 of 21

PROBLEM

Video games industry as specified by experts are a cutthroat business. The competition could be so intense that it could be categorized as win big or lose big in a single go.

Video games developer relies heavily with publishers to market and give inputs to their game. They will work hand-in-hand to ensure that the games would be a success in the market. Without publisher to give valuable inputs and marketing activities, developers would be hard to sell. Selling directly to market would also be hard, except for indie games.

The aim of this presentation is to assist for new game developer to have some insight regarding the industry, publisher, and how they correlates with users, and critics of the video games industry. Ultimately, developers also want to have some insight on any variables/clusters of variable that might resonate more with sales.

I will use some clustering model such as K-Means to get more understanding of the data. In the end, logistic regression model, and random forest would be used to make prediction.

Video Games Sales modelling

3

24/03/2023

4 of 21

DATA UNDERSTANDING

Data applied in this project is the data of video games title with sales exceeding 100,000 copies as per 30 November 2016. Some of the scheduled games after the cut-off date would appear as pre-sales.

Only the below variables would be put into consideration for our analysis:

1/ Platform = name of console where the games are resided

2/ Year = year when the games are released

3/ Genre = genre of the game

4/ Publisher = Game’s publisher

5/ Global sales = sales in number of copies

6/ Critic score = Score placed by video games critics

7/ Critic count = amount of critics evaluated the game

8/ User score = score given by the customers

9/ User count = number of customers evaluated the game

10/ Ratings = ratings given by ESBR to the game which consist of E (for all), M (mature), T (teenage), E10+ (more than 10 years old), K-A (kids to adults-retired rating), AO (adults only), EC (early childhood), RP (Rating pending)

Video Games Sales modelling

4

24/03/2023

5 of 21

HISTORICAL VIDEO GAMES SALES

Video Games Sales modelling

5

24/03/2023

6 of 21

NINTENDO HISTORICAL VIDEO GAMES TITLES

Video Games Sales modelling

6

24/03/2023

7 of 21

SONY HISTORICAL VIDEO GAMES TITLES

Video Games Sales modelling

7

24/03/2023

8 of 21

MICROSOFT HISTORICAL VIDEO GAMES TITLES

Video Games Sales modelling

8

24/03/2023

9 of 21

SEGA HISTORICAL VIDEO GAMES TITLES

Video Games Sales modelling

9

24/03/2023

10 of 21

PC HISTORICAL VIDEO GAMES TITLES

Video Games Sales modelling

10

24/03/2023

11 of 21

HISTORICAL PLATFORM VIDEO GAMES TITLES

Video Games Sales modelling

11

24/03/2023

12 of 21

GENRE OF VIDEO GAMES

Video Games Sales modelling

12

24/03/2023

There are multiple genres which are resonates with different number of sales. The genres in this presentation are as follow:

1/ Sports

2/ Racing

3/ Platform

4/ Misc

5/ Simulation

6/ Action

7/ Role-Playing

8/ Puzzle

9/ Shooter

10/ Fighting

11/ Adventure

12/ Strategy

13 of 21

GENRE AS PER NUMBER OF TITLES

Video Games Sales modelling

13

24/03/2023

14 of 21

GENRE AND SALES

Video Games Sales modelling

14

24/03/2023

15 of 21

RATING AND GLOBAL SALES

Video Games Sales modelling

15

24/03/2023

16 of 21

K-MEANS CLUSTERING - EDA

Video Games Sales modelling

16

24/03/2023

Some of uncontrolled variables such as user score, user count, critic score, and critic rating, year of games released would be clustered to classify its nature of clusters in relation to global sales. Year of the game released would also included to see the horizontal relationship between all those variables.

From the analysis we got 4 clusters with below criteria:

17 of 21

K-MEANS CLUSTERING - INTEPRETATION

Video Games Sales modelling

17

24/03/2023

Cluster 0: Cluster with minimal sales, and negative critic score means this type of games does not have any critics available, so as with user score and user count. We could classify this as Leftover Games.

Cluster 1: Cluster with moderate sales but the best user score with low critic score, we could classify this as Hidden Gem.

Cluster 2: Cluster with highest sales, good critic score and good user score, we could classify this as Successful Games.

Cluster 3: Cluster with great sales, good critic score but bad reviews from users, we could classify this as Disappointing Games.

18 of 21

MODELLING

Video Games Sales modelling

18

24/03/2023

Modelling applies several steps:

1/ Oversampling with SMOTE

2/ Multicolinearity study

3/ Scalling and feature reduction

4/ Modelling with Random Forest

5/ Modelling with Logistic Regression

The dependent variable would be global sales which I put into binary criteria based on below/above the mean of global sales. Above the mean would be categorize as “1” or successful game, below meand would categorize as “0” as unsuccessful game.

19 of 21

MODELLING - RESULT

Video Games Sales modelling

19

24/03/2023

The modelling result are as follows:

1/ PCA reduction best fit would be 20 variables which has include variance of 82 percent of the data.

2/ Random Forest resulting in precision of 85 percent and f1_score of 82 percent.

3/ Logistic regression resulting in precision of 74 percent and f1_score of 72 percent.

20 of 21

NEXT STEP AFTER THE MODELLING

Video Games Sales modelling

20

24/03/2023

The modelling shows a good percentage of accuracy, as such we could conclude for the time being the model could work to predict the data.

So, to find the answer to our question, we should go back to clustering with K-Means. Cluster 1-3 is already above mean of global sales. However, cluster 2 is the highest sales and developers could work in more detail to pick their publisher based on the number of game published, game published in recent years, etc.

This analysis only serve as general/high-level analysis to provide game developers with some insights, and all of the decision should be evaluated in detail in accordance with each developer’s unique preferences.

21 of 21

THANK YOU

Darius Ramli

Video Games Sales modelling

21

24/03/2023