<![CDATA[WebView-based attendance application in government agencies experiences performance issues (startup time 2.33 seconds, memory 186 MB, success rate 82%) and weak verification. Previous research on face recognition still used monolithic architecture (verification time 5-7 seconds) and location validation only with fixed radius without distinguishing WFO/WFA scenarios. This study aims to develop the application to Android Native with MVVM, integrate client-server face recognition and Haversine for WFO/WFA, and measure performance improvement. The method uses the Waterfall model with one group pretest-posttest design. Testing of 7 metrics with 30 samples per metric using ADB, Logcat, and observation, as well as paired t-test (α=0.05). MVVM implementation was successful (UI conformity 98.4%, layer separation 100%). Client-side face detection achieved 91.7% accuracy with 1.7 seconds (64-74% faster). Client-server communication was efficient (compression 92.2%, total verification 1.81 seconds). Server verification achieved 96% accuracy with 100% specificity. Haversine validation achieved 0.21 m error (first data). Significant performance improvement (p<0.001): startup decreased by 33.5% (2.33→1.55 sec), login decreased by 48.0% (2.50→1.30 sec), attendance decreased by 23.3% (3.60→2.76 sec), memory decreased by 27.9% (186→134 MB), success rate increased to 94% (from 82%). The development of the application to Android Native with MVVM, integration of client-server face recognition (1.81 sec, 96%), and Haversine validation (0.21 m) succeeded in improving the performance and reliability of the attendance application significantly and measurably, as well as answering the five research gaps.]]>
