A study on the results of word guessing game based on XGBoost regression modeling

Wordle is a word-guessing game from The New York Times in which the player solves puzzles through a limited number of attempts, receiving feedback for each attempt. In this paper, we build a predictive class model and use machine learning to analyze puzzle features in this game related to the number of reported results and the number of attempts required to successfully solve the puzzle, and bring EERIE's feature data file into the model to obtain percentage predictions. Then, a K-means cluster analysis model was developed, and three different clustering categories with difficulty levels of easy, medium, and hard were obtained by using all the metric items of the number of attempts of the players to solve the puzzles as independent variables. Finally, the prediction results of the percentage of all solution attempts of EERIE were imported into the cluster analysis model, and it was concluded that the difficulty attribute of EERIE was hard. This paper provides a better direction for setting puzzles that can attract more people to participate in the game.