Unit commitment scheduling of generating units in power system depends on the predicted load demand, load trend, availability of units, generating capability of generators, minimum up/down time of the units and their initial status. Previous know-how of various utility companies, differing commitment schedules of generators can result in vast difference in total operating cost acquired. Accurate hourly and daily load forecasting holds up an important purpose in appropriate scheduling of generators. This work represents a method to carry out unit commitment by employing the medium-term load forecasting based results attained from neural network learning. Both structure learning and parameter learning processes are utilised to train the neural network. The input data is composed of historical weather sensitive data such as temperature, humidity, wind speed, time variables such as hour of the day, day type (weekday, weekend, and holiday), month of the year and hourly load demand data. For structure learning, a relative analysis on the multi-layer feed forward networks and recurrent networks has been performed. The performance of the different network topologies is judged on the basis of mean square error and training time. Parameter learning is conducted on the optimally preferred network using supervised learning and the results acquired are reported.