Select a 2-class classification problem with at least ten input attributes or features and at least 1000 labeled data points, using the datasets publicly available on the internet. Balance the data, so that you have equal numbers of data points from each class, e.g., by duplicating randomly chosen members of the minority class and adding a little random noise. Use 70% of the data for training, and 30% for testing, ensuring that both sets are balanced. Train a shallow feedforward neural network (with sigmoidal node functions and one hidden layer with twice as many nodes as the input dimensionality) using back-propagation with ADAM optimizer, while keeping track of performance on test data during the training process. Repeat the experiment ten times, each time starting with a different set of randomly initialized weights; store these initial weights for future assignments. Summarize the results using one graph, plotting the average accuracy (on the y-axis) against log(number of weight updates), for training data and for test data; the graph will hence show two curves. Also show the confusion matrix(using python).