This study proposes a method for detecting non-normative articulatory features of the retroflex voiceless fricative, which is among the last sounds to develop in Polish children's speech. The approach integrates time-based phone articulation models with SVM in a hybrid binary classification method. The novelty lies in incorporating noise-based acoustic features for improved assessment and employing statistical temporal models to enhance detection accuracy. Results show that the proposed approach achieved over 86% accuracy detection for interdental and over 90% for addental articulation patterns. Dental articulation was most distinguishable, nearing 100% accuracy. Our findings indicate that adding noise features improves classification accuracy. This research contributes to the development of sigmatism diagnostic tools tailored for preschool and early school-age children, a group for whom automated pronunciation assessment remains an open challenge.