Deep Neural Networks for Fair Market Value Pricing

Trend: Deep Learning

Introduction

In the aviation industry, determining the Fair Market Value (FMV) of assets is a complex task involving numerous variables. Usually, the Fair Market Value of an aircraft is determined by a combination of factors, including the aircraft's age, maintenance history, market cycle position, and configuration, and calculated specifically for comparison by summing the values of the aircraft's components, each known as an "Aircraft Rotable".

Traditional pricing methods use historical sales and quote data in a linear way, typically using a Linear Regression Model in order to predict the Fair Market Value of a component, a fair price which may be assigned by an analyst in order to predict what a sales person might be able to sell the component for, or a guide price for the sales team depending on how it is calculated. These linear models will struggle with non-linear relationships and high-dimensional data.

This project explores and outlines the application of Deep Neural Networks (DNNs) to enhance the accuracy of FMV estimations, and provides a template for building a pricing model using a neural network. A Deep Neural Network is a type of machine learning model that is based on the structure and function of the human brain. It is a type of artificial neural network that is designed to learn from data by simulating the way the human brain processes information. Deep Neural Networks are able to learn from data by using a process called backpropagation, which allows the model to adjust its weights in order to minimize the difference between its predictions and the actual values.

A Deep Neural Network can capture non-linear relationships in order to capture complex high dimensional variables such as market demand, historical elasticity, and inventory scarcity, which a linear model cannot do.

Methodology

We propose a multi-layered perceptron (MLP) model that takes in various aircraft parameters—age, maintenance history, market cycle position, and configuration—to predict the current market value.

(What is a Deep Neural Network?)

(Where does the data come from?)

Data Preprocessing

Before feeding data into the neural network, it must be normalized. Here is a snippet demonstrating how we handle categorical variables like 'Aircraft Type' using one-hot encoding:

import pandas as pd
from sklearn.preprocessing import StandardScaler, OneHotEncoder

def preprocess_data(df):
    """
    Preprocesses the aircraft data for the neural network.
    """
    # Separate numerical and categorical features
    numerical_features = ['age', 'total_hours', 'cycles']
    categorical_features = ['aircraft_type', 'engine_model']

    # Normalize numerical features
    scaler = StandardScaler()
    df[numerical_features] = scaler.fit_transform(df[numerical_features])

    # One-hot encode categorical features
    encoder = OneHotEncoder(sparse=False)
    encoded_cats = encoder.fit_transform(df[categorical_features])

    # Combine back into a single dataframe
    df_processed = pd.concat([
        df[numerical_features], 
        pd.DataFrame(encoded_cats)
    ], axis=1)

    return df_processed

The Model Framework

We utilize TensorFlow/Keras to build the model. The architecture consists of an input layer, three hidden layers with ReLU activation, and a final output layer for the predicted price.

import tensorflow as tf
from tensorflow.keras import layers, models

def build_fmv_model(input_shape):
    """
    Constructs a Deep Neural Network for FMV prediction.
    """
    model = models.Sequential([
        # Input Layer
        layers.InputLayer(input_shape=input_shape),

        # Hidden Layers
        layers.Dense(128, activation='relu'),
        layers.Dropout(0.2), # Prevent overfitting
        layers.Dense(64, activation='relu'),
        layers.Dense(32, activation='relu'),

        # Output Layer (Linear activation for regression)
        layers.Dense(1, activation='linear')
    ])

    model.compile(
        optimizer='adam',
        loss='mean_squared_error',
        metrics=['mae'] # Mean Absolute Error
    )

    return model

Results & Visualizations

(Placeholder for Training Loss Graph)

(Placeholder for Predicted vs Actual Valuation)

Future Work

This model is an open-source tool which is open source, free to use and the details can be found on GitHub. Any templates used here are my own creation and based around mock generated data randomly generated from list pricing data and are not real aircraft values.

For more information or to contribute to this project, please visit the Contact Page.