A three-dimensional fine-segmented sampling calorimeter enables us to measure the profiles of generated shower particles along the photon’s direction. For a feasibility study, a toy detector was designed via Geant4 simulation. It consists of alternating layers of a 1-mm-thick lead absorber and a 5-mm-thick plastic scintillator. The plastic scintillator is segmented into 15-mm-wide strips, alternately oriented in the vertical and horizontal directions. The energy deposits of each strip are used to train the machine learning algorithm (XGboost) to deduce the given angle, and the resolution of its angle reconstruction is expected to be 1.3 degrees for 1 GeV photon. We fabricate a small sampling calorimeter to validate the simulation results. We use 0.15-mm-thick tungsten strips instead of lead plates and 1mm-square scintillating fibers instead of plastic scintillators for better energy resolution. This updated configuration indicates no significant difference in the angular resolution, while the energy resolution significantly improves. Its performance shows a reasonable agreement between the Monte Carlo expectation and the obtained data with a positron beam. A detailed study is underway to understand the measured data thoroughly.