Walking is one of the most unsafe activities among those that are frequently performed in daily life, as it is associated with the risk of falling. Therefore, it might be clinically relevant to evaluate and recognize the motor patterns associated with walking. Frameworks for pathological gait recognition (PGR) may support home-based rehabilitation by assessing the subject’s locomotor behavior. PGR is typically based on the recording of simulated impaired walking patterns, such as those of patients affected by neuromotor disorders (e.g., hemiplegia, ataxia, or Parkinson’s disease). This work is aimed to evaluate a Deep Learning model addressing PGR through inertial measurement unit (IMU) data. More in detail, five IMUs were placed on the human pelvises, wrists, and sternum in order to acquire raw kinematic data about the motor execution. Afterwards, inertial data are given as input to a custom convolutional neural network for learning motor patterns and classifying the human locomotion. The performance of this architecture proved the potential of the proposed workflow to recognize actual pathological gaits by exploiting simulated abnormal patterns.