I am training the Artifical Neural Network with dataset comprising of 5 numerical features. However, dataset is very small hardly less than 100 training examples. Due to which ANN model badly overfits the data. I tried the various combinations of hidden layers ranges from 1-3 and different number of neurons. I also used the regularization techniques i-e drop out and L2 regularization. I used the Relu as a activation function and Adam as optimization algorithm and learning rate in the range of 0.01 to 0.001. But didn't obtained satisfactory results. Now suggestions needed how to improve the accuracy and co-effecient of determination R2 score