<p>Dysarthria is a motor speech disorder caused due to the neurological impairments. This affects the intelligibility of speech, thus making it essential for accurate assessment of diagnosis and treatment of dysarthria. Conventional methods mostly rely on subjective evaluation, which can vary amongst speech-language pathologists. Hence, in this work, soft set theory is applied for severity classification of dysarthric speech. The strength of soft sets being handling the imprecision, and uncertainty is used, as it matches with the characteristics of dysarthric speech. Soft set theory is combined with the Mel-frequency Cepstral coefficient feature extraction and machine learning technique to build a robust classifier. The model is trained and validated on the UASPEECH dataset of annotated dysarthric speech digit samples. The results of the proposed method are compared with support vector machine, gradient boost, and random forest classifier models. Based on the proposed approach, it is evident that it improves the objectivity of assessment of severity, showing a better solution for efficient treatment method for people affected with dysarthria.</p>