This work focuses on the identification of Kokborok words (mainly the numbers or numerals) using machine learning techniques, aiming to contribute to the preservation and digital representation of this indigenous language spoken primarily in the Indian state of Tripura. A close study on the Kokborok language has revealed to have 6 vowel letters (a, e, i, o, u, w) and 21 consonants. To achieve this objective, advanced acoustic feature extraction methods have been employed, including Mel-Frequency Cepstral Coefficients (MFCC) and Perceptual Linear Prediction (PLP). These features, which capture various aspects of the speech signal, are essential for accurate and efficient speech recognition. The dataset comprised recordings from native speakers (both male and female variants of different age groups), producing a comprehensive set of Kokborok words and numbers. The audio recordings were processed using PRAAT to extract formant frequencies (F1, F2, F3), intensity, minimum and maximum intensity values, and the feature extraction techniques. The performances of these features were evaluated applying various algorithms and tuning their parameters to optimize the results. Initial findings indicated the need for normalization and feature scaling to improve model performance. Further analysis and experimentation with different models and ensemble methods were conducted to enhance the accuracy and reliability of Kokborok word (here mainly numeral) identification. This work advances the technical methodologies in speech recognition for lesser-studied languages but also serves as a significant step toward preserving the linguistic diversity of the region. By integrating modern machine learning approaches with linguistic research, this work aims to create a reliable framework for the digital preservation of the Kokborok language, ensuring its continued usage and recognition in the digital age.