Multi Layer Perceptron (MLP)

A multilayer perceptron (MLP) is a feedforward neural network with one or more hidden layers of nonlinear units that learns to map input features to a continuous or categorical outcome.

다층퍼셉트론(MLP)은 비선형 단위로 된 하나 이상의 은닉층을 가진 순방향 신경망으로, 입력 특성을 연속형 또는 범주형 결과로 대응시키는 관계를 학습한다.

How do I use a neural network to predict an outcome from tabular data?

표 형식 데이터에서 신경망으로 결과를 어떻게 예측하나요?

A multilayer perceptron is a fully connected neural network that learns nonlinear relationships between input features and an outcome through hidden layers. It fits classification or regression on scaled tabular data with enough labelled examples, though gradient boosting is often competitive on such data.

When to use Multi Layer Perceptron (MLP)

• Labelled tabular data with a continuous or categorical outcome
• Nonlinear relationships and interactions among many numeric features
• A neural component that will be combined with other network inputs

When to use something else

• Small tabular datasets or heavy use of categorical features; use gradient boosting such as CatBoost
• Coefficient-level interpretation is required; use linear or logistic regression

Assumptions and requirements

• Standardized or normalized input features
• Sufficient labelled training data
• Validation data for early stopping and tuning
• Regularization such as dropout or weight decay

What the analysis reports

• Accuracy, AUC, RMSE or similar on held-out data
• Training and validation loss curves
• Network architecture and hyperparameters
• Permutation importance or SHAP values

How to report it

The multilayer perceptron achieved a cross-validated AUC of [value] (SD [value]).

Related methods

• Logistic Regression
• CatBoost
• Random Forest

Run Multi Layer Perceptron (MLP) on Quorant