02801nam a2200253 i 4500003000500000005001700005008004100022040002500063100003600088245020900124300001600333336002100349337002500370338002300395502009800418504003900516520177200555700003202327700003702359720003802396720004302434720003302477720003702510CSPC20260817162335.0260817s2025    ph a     b    000 0 eng d  aCSPCbengcCSPCerda1 aPrades, Angela May L.,eauthor.10aRecalibrating the proxy means test model of the pantawid pamilyang pilipino program (4Ps) using ai-powered lean six sigma methodology / cAngela May L. Prades, Jessica D. Relos, and Jaselle Ann R. Lanuzo.  a133 leaves.  2rdacontentatext  2rdamediaaunmediated  2rdacarrieravolume  aUndergraduate Theses (BPA) - Public Administration.cCamarines Sur Polytechnic Colleges, 2025  aIncludes bibliographic references.3 aThe Pantawid Pamilyang Pilipino Program (4Ps) has traditionally used a Proxy Means Test, integrating with machine learning-based classification for its selection process that focuses mostly on proxy indicators for income to be eligible for membership. However, relying only on income-based proxy indicators has often led to misclassification. The main issue is that the current model does not consider non-income proxy indicators of poverty since some truly poor families are left out while others who do not qualify are included. These include not just economic conditions, but also education, health, geographic challenges, social disadvantages, and psychological well-being. This study aimed to improve the eligibility model by using the DMAIC framework to build a more accurate and data-driven classification system. The data consisted of interview transcripts in Bikol-Rinconada and Tagalog, which provided detailed scenarios about the daily lives of the vulnerable households. Thematic analysis was used to identify common patterns, and Latent Dirichlet Allocation was applied to extract probabilistic themes from the text. Rule-based scenarios were then developed to reflect the realities of the vulnerable households. These scenarios were utilized to train and test a Naive Bayes classifier. The model, called EquiSelect Classifier, was evaluated using a confusion matrix and achieved an F1 score of 82.35 percent. To further strengthen the approach, future plans include integrating real-time data, expanding the dataset to cover more regions,  xvii  and applying cross-validation techniques. The study demonstrates how transforming interview transcripts into probabilistic classification may improve targeting accuracy in conditional cash transfer programs.1 aRelos, Jessica D.,eauthor.1 aLanuzo, Jaselle Ann R.,eauthor.  aMagistrado, Jeffrey B.,eadviser.  aBarcinilla, Ejay F.,echair, panelist.  aEclar, Eugene N.,epanelist.  aArevalo, Al Lexus P.,epanelist.