Robust detection and classification of multimodal self-mixing (SM) signals emanating from the optical feedback-based SM interferometric laser sensor are necessary for accurate retrieval of sensing information. An abrupt shift in the modality of SM signals can occur due to different operating conditions; consequently, an unidentified modality shift can cause severe measurement errors. Therefore, it is necessary to detect and identify the type of multimodality so that relevant adjustments could then be made, either in the SM sensor setup or in the relevant signal processing, to avoid the errors caused by the shift in modality. In this work, SM modality identification and classification techniques based on the machine learning classifier algorithms of linear regression, XGB regressor, and decision tree regressor are proposed. The distinguishing feature values, which are used to train and test the classifiers, are extracted from the given SM signal by applying techniques such as principal component analysis, peak width, and linear discriminant analysis. Proposed methods are tested on an SM signal dataset containing a total of 45 unseen SM signals, acquired experimentally from the SM sensor. The identification and classification accuracy of the three classifiers of linear regression, XGB regressor, and decision tree regressor is 76%, 96%, and 100%, respectively.