To require the order of last mile and the increasing demands of high quality of internet, the network architecture of FTTH (Fiber to the Home) has been chosen by numerous ISP (Internet Service Provider). Poor planning does not only increase the infrastructure costs, but it also increases maintenance costs. In this study, the authors focus on the design of passive optical network by using multiclass classification [9] in backpropagation neural network to shorten the FTTH network planning based on passive optical network design to determine final splitter type which refers to the feasibility of QoS (Quality of Service) and Cost Efficiency. Dataset in this study utilizes GIS (Geographic Information System) report with 33 sub-districts in Malang Regency in 2019 data layer. As a result, after 7300 epoch, the accuracy training was 99.99% and the splitter classification accuracy was 98.76%.