1. Introduction

As an important part of the modern entertainment industry, the game type, content and experience are constantly enriched and enhanced, and its market scale and influence grow year by year. In this paper, through data analysis, in-depth excavation of player behaviour, market demand and other information for game development and sales to provide a basis for decision-making, and through predictive analysis, to help enterprises better grasp the market dynamics and layout in advance, which can help enterprises to reasonably plan their resources, reduce the market risk, and improve their economic efficiency [1]. At the same time, it can also help enterprises better meet the needs of players and enhance user satisfaction.

In terms of market trends, scholars today are committed to predicting the future development of the video game market through historical data and the current market situation [2]. As an illustration, relevant reports point out that China's game market is steadily growing in terms of both revenue and user size, with mobile games becoming a major source of revenue. Such research helps industry players to formulate more effective market strategies. As for the research on user behaviour, scholars focus on players' motivations, preferences, participation styles and the impact of game experience. The players' favourite game genres and game platforms are selected. These studies help improve game design and enhance users’ experience [3,4].

Currently, KNeighborsRegressork (KNN), Random Forest, Support Vector Machine (SVM), Decision Tree and other research methods are commonly used to predict the market of video games globally. Vikas Ukani used the Random forest model to analyse the relationship between time and global sales in his article "Video Game Sales EDA, Visualizations, ML Models"; JuHyeon_ in his article "Video Game Sales EDA, Visualizations, ML Models". Video Game Sales EDA, Visualizations, ML Models", Vikas Ukani in his article "Video Game Sales EDA, Visualizations, ML Models" used a Random forest model to analyse the link between time and global sales; JuHyeon_ in his article "Video game sales EDA&ML Classification", JuHyeon_ used Decision Tree model to analyse the relationship between time and game type, and Yonatan Rabinovich compared the relationship between time and game type in his "Video games sales Yonatan Rabinovich in "Video games sales regression techniques" compares K-Nearest Neighbors, Linear Regression, and Decision Tree models and concludes that K-Nearest Neighbor is the best [5,6].

Based on the in-depth study of the above literature, this paper aims to provide a more effective and accurate predictive analysis of the future market of video games by using multiple models to predict. By selecting appropriate datasets from the Kaggle website, data analysis and data clarity are carried out, followed by data visualization, which links the prediction of the video game market with specifics such as game genres, game publishers, etc. Then the dataset is predicted and interpreted by multiple regression models, and the one with the best accuracy and interpretation ability is selected.

2. Data pre-processing and data cleaning

The dataset for this paper is video game sales data. The data was obtained from the Kaggle webpage (https://www.kaggle.com/datasets/gregorut/videogamesales). This dataset contains the sales data of video games released from 1980 to 2020, with 16598 observations and 11 fields. the specific dataset variables and contents are as follows Table 1.

Table 1. Data set variables and types

In this paper, by importing the data and then searching carefully, it is found that there are column fields that are useful and need not be deleted. In addition, the column fields are converted into Chinese form to increase readability. In the missing value processing process, we can find that the total number of rows of the data set is 16598, of which the total number of null values is 307, the year of issue and the issuer of missing values, due to the missing data for the non-numerical type, can not be used to the average, median and other data cleaning methods to make up for the percentage of 1.8%, so choose to delete the missing value, delete the missing value, there is no significant change in the overall after deleting the missing value 17,8t data type. There is no abnormal data in this dataset, so the outlier processing is not required [7,8]. Eventually, the cleaned data is exported to CSV format, which is convenient for visual analysis by Power BI [9,10].

3. Results

In this research, this paper uses several different models for prediction, namely Linear Regression, KNeighborsRegressor, Support Vector Machine, Random Forest Regression, and Neural Networks.

3.1. Linear regression model

In the linear regression model, the data points in Figure 1 are mainly clustered around a straight line, which indicates that there is a strong linear relationship between actual sales and predicted sales. This further confirms the effectiveness of the linear regression model in predicting sales. Although some data points show large errors, most data points have small deviations. This may be because of some external factors or outliers. In order to improve the prediction accuracy, one may consider correcting for these outliers or using a more sophisticated model for prediction.

Figure 1. Linear regression model

3.2. A k-nearest neighbour regression model

As shown in Figure 2, in the k-nearest-neighbour regression model, the solid blue line runs through all the data points, showing a clear upward trend. This line represents the result of the regression analysis of the prediction model based on the data from the k nearest neighbours. It can be seen that as the predicted values increase, the actual sales have an increasing trend.

The dotted line represents the margin of error between the predicted and actual values. As can be seen from the figure, most of the data points are located on both sides of the dotted line, indicating that the difference between the predicted and actual values is relatively small. This further proves the accuracy and reliability of the K nearest neighbour regression model.

Figure 2. K-nearest neighbour regression model diagram

3.3. SVM

In the SVM model, Figure 3 shows how the SVM regression model improves its prediction accuracy as actual global sales increase. The blue dotted line represents the predicted value, while the light blue area represents the 95% confidence interval. It can be seen that as sales increase, the predicted values gradually approach the actual values, which indicates that the SVM regression model has a good fitting ability.

Figure 3. Support vector machine model

3.4. Random Forest Regression Model

Similar to SVM regression, the below graph shows the prediction accuracy of the Random Forest Regression Model improves as actual global sales increase (Figure 4). However, because Random Forest uses multiple decision trees for integrated learning, its predictions may be more accurate than a single SVM.

Figure 4. Random forest regression model

3.5. Decision Tree Regression Models

In the decision tree regression model, the predicted values are progressively closer to the actual values as the actual global sales increase (Figure 5). However, decision tree regression may not be as accurate as other models in interpreting forecasts.

Figure 5. Decision tree regression model

3.6. Neural network prediction model

In the neural network forecasting model, a clear linear trend can be observed, i.e., as the actual global sales increase, the predicted global sales also increase. This indicates that the neural network model can capture the trend of sales better.

In Figure 6, it can be seen that there is a certain deviation between the predicted and actual values, but overall a better fit is presented.

Figure 6. Neural network prediction model

4. Model comparison and analysis

This paper compares and analyses the results of the above models in terms of mean square error and R-squared. The linear regression model performs best in mean square error but has lower R-squared. The decision tree model has a higher mean square error but also a higher R-squared compared with linear regression. The Support Vector Machine model had a higher mean square error and R-squared, indicating that the model may be overfitting.

The K Nearest Neighbour model has a mean square error and R-squared close to zero, suggesting that the model may not fit the data well. The random forest model had the lowest mean square error but the highest R-squared, suggesting that the model may have good explanatory power. To summarize, the linear regression model performed best in terms of mean square error while the random forest model performed best in terms of R-squared. This implies that the linear regression model outperformed the other models in terms of predictive accuracy while the random forest model outperformed the other models in terms of explanatory power.

5. Conclusion

This paper uses a linear regression model for video game sales analysis and forecasting In conclusion, this EDA will help us gain insight into the video game sales dataset. By exploring the data and visualizing the relationships between different variables, we can understand patterns and trends in the gaming industry. In addition, building machine learning models will allow us to predict global sales with great features. The Video Game Sales dataset provides valuable information about video games, including their sales in different regions, genres, publishers, platforms, release years, and global sales. By exploring and analyzing this dataset, we could have a more profound understanding of patterns, trends and preferences in the gaming industry. Additionally, correlation heatmaps can help us identify the relationships between different variables and their impact on global sales. With machine learning models, we can predict global sales based on the available features to further our understanding of the factors that influence the sales of video games.

Authors Contribution

All the authors contributed equally and their names were listed in alphabetical order.