A Targeted Assistance Screening System Using Decision Tree Technique

Abstract

Poverty remains a critical challenge in developing countries, particularly in northeastern Thailand, where the proportion of poor households is relatively high. Traditional screening systems such as TPMAP still lack accuracy and mainly rely on economic indicators, overlooking social and cultural dimensions, which leads to misallocation of resources and failure to reach the intended target groups. Therefore, this study had two main objectives 1) to develop a decision tree model for multidimensional poverty screening and extract IF–THEN rules from household data, and 2) to evaluate the model’s performance in comparison with the existing TPMAP criteria and analyze the logical rules derived from the model. The sample consisted of 1,766 households in Sisaket Province, with data obtained from field surveys and official government databases. The research instruments included a household questionnaire, the Sisaket Equity System (SES) database, and the developed decision tree model. Descriptive statistics (percentage, mean, and standard deviation) were employed, while model performance was evaluated using 5-fold cross-validation with Accuracy, Precision, Recall, F1-score, AUC, and the confusion matrix as evaluation metrics.

 The findings revealed that 1) the developed decision tree model achieved an average accuracy of 0.82 and an average AUC of 0.79, statistically outperforming the TPMAP criteria across all evaluation metrics; and 2) the model was able to identify target households more accurately and in an interpretable manner. The extracted IF–THEN rules indicated that social capital and human capital were the most influential factors in classifying poverty status, followed by physical, financial, and natural capital.