In recent decades, Earthquake Early Warning (EEW) has become an option in many countries to reduce the vulnerability to earthquake hazards. P-wave picking under noisy conditions is challenging; the accuracy of P-wave picking affects the accuracy of Peak Ground Acceleration (PGA) in the warnings issued by the EEW system. Western Java, Indonesia's most densely populated region, has a massive infrastructure development. The lack of EEW in Indonesia, especially in western Java, makes it challenging to provide EEW to the community and end users. We employed an integrated method approach to developing onsite EEW in the western Java region, aiming to improve the lead time and accuracy of earthquake information, particularly in estimating PGA. We combined PGA estimation using Support Vector Regression (SVR) and validated it against linear regression estimates to assess its reliability across 5706 earthquake signals. This combination yields more accurate PGA estimates with high confidence when public earthquake warnings are issued. We used six features extracted from the first 3 s of the P-wave to build an SVR model for predicting PGA in the EEW system. The six parameters are Pa, Pv, Pd, τc, IV2, and CAV, which are directly correlated with PGA, magnitude, and earthquake damage potential. From an investigation of 485 earthquake events in western Java, Indonesia, that occurred between 2009 and 2023, with varying magnitudes, we obtained a threshold for the average period ðτcÞ of 0.55 s and a peak displacement (pd) of 0.4 cm. PGA estimates from linear regression and machine learning approaches are pretty accurate for two damaging earthquakes: the 2009 Tasikmalaya earthquake (M 7.3) is 636.9 gal, and the 2022 Cianjur earthquake (M 5.6) is 483.6 gal. The use of the IPP (Initial Power of the P-wave) method provides better results than the Short Time/Long Time Average (STA/LTA), InTeg, and EQTransformer methods. Based on the study results, it is necessary to develop and implement onsite EEW using an integrated approach in Western Java, with adequate acquisition instruments and communication systems. #c World Scientific Publishing Company.