About this Model

The model structure is a multi-layer perceptron, the simplest

form of a feed-forward neural network. Because the neural net was

coded in Rustlang, it was coded from scratch. The linear algebra

for the neural net is the same as a normal multi-layer perceptron

with an input layer, a middle layer, and an output layer. Each layer

is made up of simulated neurons, which use an activation function

and weighted sum to simulate the 'action potential' firing

of their organic counterparts. The activation function for

each layer of the network is different. The first layer uses a

hyperbolic tangent as the activation function, the second uses

a logistic function, and the third uses a 'step function' to

consolidate the values into a 1/0 output. The outputs of each

layer's activation function become the inputs of the next layer

and are hence forward propagated to generate an output. This

output is then compared to an inputted expected output. The

difference between the expected and actual output is then used

to adjust the weights of each neuron, so it fires more accurately.

The data is based on Australian weather readings, so the

prediction is valid for that geographic area alone. A lot

of imputation was done on the data during preprocessing.

Note that the dataset is imbalanced so the trained model is

biased towards predicting it will not rain tomorrow. This

is because the model is aMulti-Layer Perceptron trained

on a relatively small amount of data. With more data, this

bias could be trained out of the model, however, only 15,000

rows of data were used to train the model. If there were around

100,000 rows of data, then the model would be more accurate.

With 15,000 rows, there are some techniques that could have

been used to remove this bias, but it runs the risk of biasing

the model towards the raining condition, due to the small amount

of data. The reason an MLP was used as opposed to a Logistic

Regression Model or Bayesian network is because this is a

proof of concept for hosting prediction models using the blockchain.