Development of Inter-Island Sailing Safety System Through Machine Learning with IoT-Based Data Integration

Abstract

Abstract—Unpredictable weather phenomena pose significant risks to maritime vessels navigating inter-island waterways near Belakang Padang, Batam, particularly for smaller craft. While meteorological services provide regional forecasts, these macroscopic datasets frequently omit localized anomalies critical to navigation safety at particular waypoints. Addressing this information gap, the research proposes an IoT framework enabling continuous, hyper-localized weather parameter measurement from two geographically-separated monitoring stations. The system architecture comprises two separate units: an onboard Node mounted on the vessel and a shoreside Station. These components communicate using Long Range (LoRa) technology and implement the K-Nearest Neighbor (KNN) classification method to assess sailing feasibility. The Node subsystem focuses on vessel dynamics by measuring tilt angles, while the Station subsystem concurrently captures environmental data specifically temperature, humidity, wind velocity, and precipitation which are displayed on a real-time web interface. Testing confirmed that LoRa communication succeeded across distances reaching 8.67 km. Linear regression calibration of the temperature sensor yielded a high-fidelity model (R² = 0.9998), reducing measurement error from an initial 4.04% to 2.88%. Using K = 7 and Manhattan distance weighting, the KNN model achieved 98.90% classification accuracy with an average 10-fold cross- validation score of 96.72%. These results demonstrate the platform's capability to monitor maritime weather patterns continuously and categorize seaworthiness with high precision, thereby offering substantial value for reducing shipping incidents. Keywords: IoT, LoRa, K-Nearest Neighbor, maritime weather monitoring, seaworthiness.