How can we redirect partisans from the escalating partisan attachment (i.e., partisan strength) fuelling the surge of affective polarization in America and the UK? Despite ongoing empirical research identifying predictors of partisan strength, there's a gap in understanding the influence of non-theory-driven factors on its increase. To uncover predictors of rising partisan strength and inspire novel hypotheses, we embraced a data-driven approach. Utilizing multiple extreme gradient boosting trees (XGB) and elastic net linear models, we predicted changes in partisan strength among British Election Study respondents from wave 7 to wave 9, based on their responses to 105 other items. The linear and non-linear models revealed distinct sets of predictors for the growth in partisan strength. Non-linear XGB models pinpointed social identity and affect-based items (“pidConnected”, “euID4”, “pid_scale”) as primary predictors, whereas linear elastic net models highlighted issue-based items (“efficacyNoMatter”, “bestOnMII”, “LRAL_mii_cat”). Our findings indicate that issue-based attitudes' impact on the growth of partisan strength aligns more with a linear relationship, while the effects of social identity or affect-based attitudes are more accurately captured non-linearly. We are currently replicating this study using the UK Household Longitudinal Study panel waves 10 and 12, and BES waves 21 and 23, following the same methodology. Our preliminary results illuminate underexplored dimensions of partisan strength from theory-driven perspectives, offer deeper insights into how the mechanisms' linearity varies between political, psychological, and social attitudes impacting partisan strength, and support further differentiation in studying social identity and issue attitudes in partisan strength research.