Optimization Method for Predictive Models Based on ARIMA Time Series and K-means Clustering Algorithm

This paper proposes a method for optimizing prediction models based on the ARIMA time series and the K-means clustering algorithm to address the challenge of solving world puzzle games. The method aims to offer players a diverse range of challenges by anticipating and predicting imbalances in difficulty and content in advance. To achieve this, the paper first establishes an ARIMA time series model to fit the data and calculate the goodness of fit. Subsequently, employing keywords such as vowels, consonants, word composition, and usage, the K-Means clustering algorithm is applied to compute the clustering center of the data and generate a clustering result graph. Finally, different regression models were established for the characteristics of various difficulty samples, and the words were digitally encoded. The least squares method was used for fitting. Faced with different attempts, a reasonable prediction interval was obtained based on the analysis results. Experimental results indicate that, in comparison to traditional algorithms, the proposed method demonstrates higher accuracy in predicting word attribute classifications. The findings of this research offer an effective approach for game developers to manage puzzle game difficulty, enabling better balancing of game challenges and enhancing the gaming experience and engagement for players.