Floods are a natural phenomenon that frequently occurs in Indonesia, including in Samarinda City which has faced flood issues over the past three years, affecting thousands of homes and around 27,000 residents. Predicting flood disasters requires machine learning technology using data mining classification methods. However, classification processes often encounter issues related to high-dimensional data, which can lead to overfitting and class imbalance, thereby biasing dominant classes while neglecting minority classes. This research aims to enhance classification accuracy in Samarinda City's flood data using the K-Nearest Neighbor (KNN) algorithm combined with Relief feature selection and Particle Swarm Optimization (PSO) optimization. The validation method employed is 10-fold cross-validation, with performance evaluation using a confusion matrix. Data sourced from Samarinda City's Disaster Management Agency (BPBD) and Meteorology, Climatology, and Geophysics Agency (BMKG) spans from 2021 to 2023, comprising 19 features and a total of 1095 records. Relief feature selection identified four crucial features: maximum wind direction, wind speed, average wind speed, and maximum wind speed direction. Average evaluations with k values of 3, 5, 7, 11, 13, and 15 demonstrate that Relief feature selection and PSO optimization effectively enhance accuracy in the K-Nearest Neighbor algorithm for flood data, with KNN and PSO yielding improvements of 2-5%. Relief feature selection alone improves accuracy by 1-2%, while combining Relief with PSO provides a 2-5% enhancement. The combined KNN, Relief, PSO model is expected to deliver optimal performance in classifying Samarinda City's flood data.