Particle identification (PID) was developed for a newly constructed \(4\pi \) CsI(Tl) detector array using machine learning methods. This work presents the construction of the CsI (Tl) array, experimental setup, and the machine learning approaches employing fuzzy c-means (FCM) and support vector machine (SVM). Unlike traditional fitting methods which are typically effective only at high energies, the FCM and SVM provide robust particle identification across the full energy spectrum, thereby enhancing data quality and counting statistics in fusion evaporation experiments. Particles such as \(\upalpha \) (Z = 2), \(\text {p}\) , \(\text {d}\) and \(\text {t}\) (Z = 1), as well as \(\upgamma \) (Z = 0) can be classified automatically using FCM. With dedicated training on five distinct waveforms corresponding to particles emitted from fusion evaporation reactions, the SVM demonstrates enhanced separation capability. The SVM outperforms FCM, particularly in scenarios involving weakly bound projectiles such as \(^{7}\text {Li}\) or \(^{9}\text {Be}\) . The effectiveness of these classification models was evaluated through offline analysis using \(\upgamma \) -ray coincidence and particle-gating techniques. The results demonstrate that machine learning methods provide comprehensive particle identification across all energy ranges and the SVM method contributes to an approximate 50 \(\%\) increase in particle-gated coincidence statistics in the present experiment. The integration of machine learning in particle identification opens new possibilities for advancing nuclear-structure studies and analysis of complex reaction dynamics.
