Clouds are one of the most important factors in climate and weather changes, and A ceilometer is used to automatically observe information about cloud altitude and cloudiness. In this paper, we propose a method to detect the presence-or-not of meteorological phenomena by applying machine learning to the back-scattered data collected from the ceilometer. First, to eliminate the noise in the observation data, linear interpolation and denoising autoencoder is used to perform noise elimination on the back-scattered data. Since the meteorological phenomena is remarkably small, the machine learning method was applied after undersampling. The machine learning used are Random Forests, Support Vector Machine, and Artificial Neural Network. In case of support vector machine and artificial neural network due to learning time problems, experiments were performed by reducing learning factors by feature selection. We deal with precipitation, fog, and atmospheric pressure as meteorological phenomena using AWS and visibility data. The F1-score was used as an accuracy measure for the performance evaluation of machine learning. In the experiment, it is 0.3377 for the precipitation, 0.0949 for the fog, and 0.3494 for the atmospheric pressure.