The most important investment decision when it comes to implementing photovoltaic (PV) array(s) either for commercial or private application, is the levelized cost of energy (LCOE) calculation. Whereas LCOE, used as an assessment tool, to calculate the cost effectiveness of energy generation in relation to the return over investment in a prescribed time, can be calculated with a simple mathematical equation; it remains unattainable without a proper systems dynamic modelling and performance prediction of PV modules in the array(s). PV cells are unpredictable and have very low conversion efficiency of about 15–25% which makes it often necessary that maximum power point tracking (MPPT) system is integrated to PV modules to get optimum energy yield. This draw back, makes it quite crucial that output power and performance prediction of PV arrays at module level are established as a precursor to PV solar system design and implementation. This is key to sustaining efficient energy production by the PV generators, as it helps to detect if the DC power output of the PV modules is at optimum. In reality; it is quite challenging to accurately predict the performance of PV array power output due to variable factors like solar insolation, sun incident angle, temperature, PV array configuration, dust and non-uniform illumination amongst many factors. All these factors create nonlinear output characteristics in PV modules which results into instability in energy yield, faults, losses and costly maintenance of the PV infrastructure. This research study presents a method for the performance prediction of PV modules in an array. It involves using neural network model with three layers (input, hidden and output) to predict PV module energy yield under uniform and non-uniform illumination (shaded conditions). Typical data like open circuit voltage (Voc), short circuit current (Isc), solar insolation and cell temperature are taken into consideration. Training datasets are obtained from the PV under investigation and applied to the neural network using backpropagation algorithm to train the datasets. The PV module(array) maximum output power performance evaluation is verified by comparing the predicted neural network output power with the empirical measurement from the PV module(array) under investigation.