I have a dataset that has 24 input parameters and 1 target parameter. The number of data is about 14800. To build a strong neural network, I tried different number of layers and number of neurons, but I failed to achieve my ideal accuracy. How should I do it؟

My target parameter is actually the efficiency value for a collector, which varies between 3 and 96%. The best error value I got was MAE = 1.5, and my goal is to reach a number below 0.5.