Identification of Mad Thabi'i Reading Errors in Surah Al-Kahfi Verses 1–10 Using Long Short-Term Memory: A Case Study of Students at DTA As-Salaam

Mad Thabi'i is the most basic elongation rule in Quran recitation; however, many students at Diniyah Takmiliyah Awaliyah (DTA) As-Salaam still have difficulty in applying it correctly, while manual evaluation by teachers is limited by time constraints. This study proposes an automated approach to identify Mad Thabi'i recitation errors using a Long Short-Term Memory (LSTM) model. A dataset consisting of 710 valid audio samples representing correct and incorrect pronunciations of the three Mad Thabi'i letters (alif, waw, and ya) was collected from teacher recordings. Audio features were extracted using Mel-Frequency Cepstral Coefficients (MFCC) with delta and delta-delta coefficients, and data augmentation was applied to improve model generalization. A stacked LSTM model containing 61,249 trainable parameters was evaluated using stratified 5-Fold Cross Validation, achieving an average validation accuracy of 84.93% with an F1 score of 0.8453. Evaluation on an independent test set of 107 samples yielded an accuracy of 88.79% and an F1 score of 0.8877. Performance analysis showed the highest accuracy for kasrah (97.3%) and the lowest for fathah (81.0%), indicating that Mad Thabi'i readings involving alif with fathah were the most frequently misclassified among the evaluated categories.