Sensory Precipitation Forecast Using Artificial Neural Networks and Decision Trees

Abstract

Meteorology stations sold in the market have various difficulties in terms of their use, also these systems are costly to obtain. With state of the art sensor technologies, the development of mini weather stations has become easier. This study focuses on the development of a model weather station device using temperature, relative humidity, UV, LDR Light, rain and soil moisture sensors to collect major environmental data. The measured data were wirelessly transmitted to the remote station for logging via the GSM module and the information was sent to the database in the internet environment. In addition, the data from the sensors are organized by correlation. The classification was made according to the data obtained from the rain sensor and the relationship between the other 5 sensors used in the device to the rain classification was examined. Sensor data were scaled between 0-1 with min-max normalization before being subjected to deep learning and machine learning training. In the Decision Tree (DT) a model score of 0.96 was obtained by choosing the maximum depth of 20. The artificial neural network (ANN) yielded a classification score of 0.92 using 4 hidden layers and 100 epochs in the artificial neural network model.