Robust vowel recognition is essential for Automatic Speech Recognition (ASR) systems, particularly in low-resource languages like Malay. This study evaluates the classification performance of Malay vowels (/a/, /e/, /o/, /u/, and /i/) using four feature sets: 13 Mel-Frequency Cepstral Coefficients (MFCCs), 33-MFCCs, 13-MFCCs with formants, and 33-MFCCs with formants. Two classifiers, Logistic Regression (LR) and Neural Networks (NNs), were assessed to determine the impact of feature dimensionality and spectral information on recognition accuracy. Results show that Neural Networks consistently outperform Logistic Regression across all feature sets, achieving the highest accuracy of 98.07% with 13-MFCCs and formants. While Logistic Regression performs competitively with simpler feature sets, it struggles with spectral ambiguities in higher-dimensional spaces. These findings emphasize the importance of integrating spectral and formant features for improved Malay vowel classification. This study offers practical insights into feature selection and model design for ASR systems in low-resource languages, paving the way for future research into hybrid and deep learning models for multilingual speech recognition.