PD detection is an effective way to evaluate the degradation state of cable insulation. The extraction and selection of relevant features from PD raw data have been mostly investigated to recognize the types of insulation defects in HV equipment. In this study, two different feature extraction methods combined with supervised classification techniques are implemented for ageing state recognition of a polyethylene-insulated cable under HVDC conditions. For this purpose, an original experimental setup is implemented. Experiments are performed on a long length 100 m coaxial cable subjected to high electric fields. PD events are detected by direct coupling and collected with a digitizing oscilloscope. Feature extraction based on PD pulse shape parameters represented in time domain as well as wavelet decomposition coefficients are used separately as input variables of Support Vector Machines classifiers (SVMs). A feature selection method is implemented to design optimized SVM classifiers that attribute an ageing state to the cable insulation. The classification performance achieved with both feature extraction methods are presented and compared. The results show satisfactory recognition rates of two ageing states of cable insulation, up to 100% with a small subset of variables, particularly when features are extracted from wavelet decomposition of PD experimental data.