LeaseLabs is a relatively small company active in the secondhand car market and car leasing industry. They have extensive domain knowledge and were looking to combine this with a more data-driven approach to their business. That’s where we as Notilyze came in.

The focus of this project was to gain insights into what factors play a role (and to what extent) in the valuation of cars in the secondhand market. Furthermore, we wanted to analyze car depreciation over time for a couple of reasons. First, seeing how certain type of cars depreciate over time can be valuable for determining the moment of purchasing/selling. Second, car depreciation is an important component of the lease price and it is therefore highly relevant to obtain good estimates.

We received data about secondhand car sales and combined this with publicly available data about car registration from the Dutch government (RDW). In the figure below, some analysis is shown on the relationship between car price and various explanatory variables.

This analysis gives some insights into which factors play a role when determining prices. Mileage (kmstand), power, original price, vehicle age – amongst others – seem to play a role. Intuitively, this makes perfect sense.

After this analysis, we dove deeper into the data and developed several machine learning models to fit the data. Examples include linear regressions and a variety of tree-based models such as decision trees, random forests, and gradient boosting. Subsequently, we tuned the parameters of the models and selected the best performing model using the well-known mean squared error (MSE) criterion.

The selected model provides an estimate of a car’s worth on the secondhand market based on some model inputs (which include but are not limited to the car features mentioned above). Since one of these features is vehicle age, we could provide some insights into car depreciation by varying this model input while keeping other variables constant. To illustrate this, some examples of car depreciation is shown in the figure below.


Applying some business rules, potentially interesting vehicles could be selected to investigate further – or keep an eye on in the future – based on their ‘depreciation course’.

Using the model to estimate car prices on the secondhand market, Notilyze developed several API services for LeaseLabs in SAS Intelligent Decisioning. One API simply returns an estimated price based on known car features, while another one uses the model and other business rules to determine lease prices. In the latter, car features are input as well, and the service returns the lease price with an overview of its components.  The figure on the right displays the development environment of this lease price API in SAS Intelligent Decisioning. Purple blocks are pieces of code, orange blocks are conditions and the yellow block represents the model.

With our analyses and developed APIs, LeaseLabs’ decisions are now not solely based on domain knowledge and gut feeling but supported by data! A great example of real-time Analytics-as-a-Service.