Machine Learning Model for Commercial Lease Benchmarking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current commercial lease pricing methods are ineffective due to reliance on human interaction and subjective judgment, leading to biased and inaccurate benchmarking and pricing results, especially for properties with multiple leases and diverse characteristics.
Innovation Solution
A machine learning-based method for commercial lease benchmarking that selects an optimal machine learning model trained on a feature dataset including property data and metric data, and uses cross-validation to ensure accuracy, allowing for improved predicted lease values with minimal information such as a property address.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If human interaction and subjective judgment are used to determine lease values, then flexibility and adaptability are maintained, but accuracy and objectivity of pricing results deteriorate due to bias
Solution Approach 1:
The patent replaces the mechanical system of human judgment and interaction with an automated machine learning model that processes property data and metric data to generate lease valuation predictions. This substitution eliminates human bias while maintaining adaptability through the model's ability to learn from diverse data features.
Solution Approach 2:
The patent introduces a machine learning model as an intermediary between raw property data and lease valuation results. This intermediary processes multiple data features including property characteristics and metric data to produce objective, accurate pricing while removing the need for direct human judgment in the valuation process.
2Measurement precision
If more information is provided in lease valuation inquiries, then accuracy of benchmarking improves, but complexity of data collection and processing increases
Solution Approach 1:
The patent creates a universal machine learning model that can process multiple types of data features (property data and various metric data) through a single integrated system. This multi-functional approach allows the system to handle diverse information requirements without proportionally increasing complexity, as the same model architecture processes different data types.
Solution Approach 2:
The patent transforms complex unstructured property information into structured metric data parameters that the machine learning model can efficiently process. By converting diverse property characteristics into standardized numerical features, the system achieves high accuracy without proportional increases in processing complexity.
3Ease of operation
If minimal property information is provided in valuation requests, then ease of operation improves, but accuracy of predicted lease values deteriorates
Solution Approach 1:
The patent performs preliminary actions by pre-training the machine learning model on extensive property data and metric data before actual lease valuation requests. This pre-processing of data and training of the model enables the system to generate accurate predictions from minimal input information, as the model has already learned relationships between various property features and lease values.
Solution Approach 2:
The patent creates a virtual copy of the complex property valuation process through the machine learning model. Instead of requiring users to provide comprehensive property information, the model replicates the expertise of professional appraisers by processing minimal input data through learned patterns, producing accurate valuation predictions without requiring extensive user-provided information.
4Measurement precision
If multiple machine learning models are trained and evaluated, then accuracy of selected model improves, but time and computational resources for model selection increase
Solution Approach 1:
The patent applies partial action by performing cross-validation on a selected subset of machine learning models rather than exhaustively evaluating all possible models. This approach achieves sufficient accuracy by testing multiple candidate models with different architectures and hyperparameters, selecting the best performing model without requiring exhaustive search of the entire model space.
Data Source
AI summary
Methods, non-transitory computer readable media, and property analysis server devices are disclosed that generate a feature dataset including property and metric data for properties includes an address and an actual lease value. A machine learning model (MLM) trained on the feature dataset is selected from different types of MLMs. A determination is made that the selected MLM exceeds an accuracy threshold based on a cross-validation using predicted lease values. The property data is stored in a lease benchmarking database with the addresses replaced in with corresponding geographic coordinates and geohash values. The properties are associated in the lease benchmarking database with the predicted lease values. One of the predicted lease values is returned in response to a received lease pricing request that includes an address. The returned predicted lease value is identified in the lease benchmarking database based on a geographic proximity of the address to one of the properties.