Authors:
Pratham Maan
1
;
Lov Kumar
2
;
Vikram Singh
2
;
Lalita Murthy
1
and
Aneesh Krishna
3
Affiliations:
1
BITS-Pilani, India
;
2
NIT kurukshetra, India
;
3
Curtin University, Australia
Keyword(s):
SMOTE, Machine Learning, Multilayer Perceptron, Video Game.
Abstract:
The video game development industry deals with all aspects of video game development, including development, distribution, and monetization. Over the past decade, video game consumption has skyrocketed and the industry has witnessed remarkable technological advances, although it has stumbled across some bottlenecks. The lack of a well-formatted game’s postmortem video is one pivotal issue. A postmortem video is published after the game’s release, to track its development and often understanding ’what went right and what went wrong’. Despite its importance, there is a minimal understanding formal structure of postmortem videos explored to identify video game development-related problems. In this work conducted a systematic analysis of the chosen video game problem dataset extracted from postmortem videos with 1035 problems. We designed Multilayer Perceptron (MLP) classifiers for early identification of video game development problems based on their description or quote. The empirical
analysis investigated the effectiveness of 09 MLP-based classification models for identifying video game development problems, using 07-word embedding techniques, 03 feature selection techniques and a class balancing technique. The experimental work confirms the higher predictive ability of MLP compared to traditional ML algorithms such as KNN, SVC, etc, with 0.86 AUC values. Moreover, the effectiveness of class balancing and feature selection techniques for selecting the best feature set is evaluated by box plot and Mean rank test using the Friedman Mean Rank test on the null hypothesis, indicating an impact on the overall predictive ability of MLP models with AUC values of 0.862.
(More)