C19 Update with Brazil and World Neural Network

2020-06-15

Total US C19 deaths 115,732

Last five daily = [921 888 846 767 296]

Future estimates

   123800 +/- 1963  =  Jul 1 total (95% confidence)

   132752 +/- 1963  =  Aug 1

   136991 +/- 1963  =  Sep 1

Total world C19 deaths 433,066

Last five daily = [5605 4395 3933 4342 3330]

Future estimates

   472188 +/- 74468  =  Jul 1 total (95% confidence)

   538839 +/- 74468  =  Aug 1

   584676 +/- 74468  =  Sep 1

Total Brazil C19 deaths 43,332

Last five daily = [1274 1239  909  892  612]

Future estimates

   59933 +/- 18737  =  Jul 1 total (95% confidence)

   83032 +/- 18737  =  Aug 1

   94861 +/- 18737  =  Sep 1

Neural network fit and prediction to world daily C19 deaths is given below. The NN uses MLPRegressor from scikit_learn. Here is the print of the neural net used to make the plot. A couple of things to note. (1) Result has strong dependence on random seed used, note random_state=2 in the printout; choosing a different random_state changes everything, nothing special about the choice of 2 except it looks reasonable to me. (2) Result is also strongly dependent on data averaging, here I use 7-day averaging which, again, looks reasonable to me. Future estimates are also trailing 7-day averages, but here they are:

Future estimates (trailing 7-day avg)

   3609  =  Jul 1 daily (day 161)

   3218  =  Aug 1 daily (day 192)

   2975  =  Sep 1 daily (day 223)

########### Neural Network #############

Score =  0.985

Neural Net: MLPRegressor(activation='relu', alpha=0.0001, batch_size='auto', beta_1=0.9,

beta_2=0.999, early_stopping=False, epsilon=1e-08,
hidden_layer_sizes=(100, 100, 100), learning_rate='constant',
learning_rate_init=0.001, max_fun=15000, max_iter=2000,
momentum=0.9, n_iter_no_change=10, nesterovs_momentum=True,
power_t=0.5, random_state=2, shuffle=True, solver='lbfgs',
tol=0.0001, validation_fraction=0.1, verbose=False,

warm_start=False)