Comparative Analysis of Crossword Puzzle Scoring Prediction Models Based on ARIMA, GBDT, and BP Neural Network

This study utilized ARIMA, GBDT, and BP neural network models to forecast the results of the “World” crossword puzzle game in 2023. By analyzing score data reported by players on Twitter, we investigated the accuracy of different models in predicting the number of word guesses and time. Results indicate that the number of consonants and lexicality have the greatest impact on scores in difficult mode. The ARIMA(1,1,1) model forecasted a result of 14,610 with a confidence interval of [8320,23875], indicating good predictive performance. This study provides a novel approach to forecasting crossword puzzle games and serves as a reference for similar game predictions.