Keywords: Clustering, Data Mining, Statistics, Modeling, PCA
While many RPG video games stimulate players’ critical thinking, their efforts can lead to wasted hours and poor outcomes. The objective of this paper is to use multivariate data mining in the RPG video game “Empire: Four Kingdoms” to optimize rank/resource outcomes. 2 main game objectives are: (1) recruit a powerful army to conquer more land & defend against enemy attacks, (2) produce new resources (Food, Wood, Stone) for building a mighty fortress. We focus on the 1st out of 5 development stages: concentration on internal affairs, enhancement of defense power, and control of food consumption. There are 40+ types of troop units available to build an army in the 1st stage. Each troop unit has 7 attributes (defense melee/ranged, attack melee/ranged, traveling speed, looting capacity, food consumption). To optimize the troop recruiting process, Multivariate Correlations are used to identify affinity patterns among all attributes and across all troop units. While correlation analysis suggests splitting troop units into 3 categories (Attack, Defense, Speed/Looting), the result is mathematically unsound. In the 1st Stage, Defense capability is more important than Attack and Speed/Looting. There is one battle constraint – allow limited troop units/types per battle – which limits the linear optimization model; Hierarchical clustering is used to resolve this by more accurately identifying clusters. Principle Component Analysis (PCA) is then used to verify dimensional reduction and determine the appropriate number of clusters (eight) with the use of a Scree Plot. We selected the most powerful troop unit from each cluster in recruiting, and then defined a Castle Power Index function based on the 1st stage objectives. The optimal recruiting setting was found based on the desired specifications. The model created is validated in actual game battles with 95% prediction interval. This statistical approach enhances troop recruiting and accurately predicts battle results.