Skip to main content

Table 2 Key features of algorithms used

From: How much do social connections matter in fundraising outcomes?

MLP

Decision tree

LR

Random forest

Feed data to input layer

Set connection factor w, offset b in hidden layer

Use Sigmoid function as activation function

Output classification results

Compute information gain for each variable

Select the variable with largest information gain as the root node

Repeat above steps until tree is built

Construct prediction function, using Eq. (3)

Construct loss function J(θ)

Using gradient descent to find smallest J(θ)

Random sampling Split nodes

Repeat step 2 until no more splits

Repeat steps 1 to 3 to build decision trees to form a random forest

Bayesian inference

KNN

AdaBoost

SVM

Set a priori probability

Set conditional probability from the given information

Transform a priori probability into posterior probability with the information

Randomly select n nodes as classification centres

Compute distance from each node to the centres

Update centres

Repeat steps 1 to 3 until convergence

Initialize weight distribution D1

Train weak classifier

Combine the trained weak classifiers into a strong classifier

Find number of classifiers

Apply kernel function

Train data to obtain hyperplane

Repeat steps 1 to 3 until convergence