Chapter 8 — Assessment
8.4. British Airways Cargo Consignment Problem
The main aim of cargo consignment is to maximise net profit by accepting the ideal types of cargo requests, prior to a aircraft taking off. Therefore, the problem that most airlines have to solve is when to accept or reject a potential cargo request. The acceptance or rejection of cargo depends on many pieces of qualitative and quantitative information which can cause a real decision problem for the human sales agent who has to make the final decision. This was the main reason that British Airways had devised the ASP algorithm (Acceptable Shipment Package) [37] to assist the decision making process.
As described in Chapter 6, the ASP algorithm is a simple weight-scoring mechanism that implements a point scoring system for each cargo request, based on factors such as the consignment weight and volume, density of consignment, rate offered and the urgency of the consignment. However, the weighting given to these various factors has been hard to fine-tune, with the result that cargo yield analysts often disagree with the ASP recommendation when a request is scored in the range 90 to 110 (where 100 is the threshold for accepting a booking). They often decide to accept bookings which ASP would reject because the flight is not filling as quickly as the usual trend. They may also reject the ASP accept recommendations if the booking requirement does not correspond with what is available.
The hybrid solution to the cargo consignment problem, as described in Chapter 6, uses neural networks trained on historical cargo acceptances and rejections, combined with an improved version of the ASP algorithm, to make the final accept or reject decision. The hybrid solution was run against a test set of previously unseen cargo requests for Auckland, Melbourne and Sydney. This same test set was also run against an implementation of the British Airways ASP algorithm and against the improved ASP algorithm used within the hybrid solution. Table 8.1 shows the results for these three systems.
These results clearly show that the hybrid solution produces a better quality result than the existing purely expert system based approach (i.e. ASP) to solving the cargo consignment problem. The hybrid solution classified 21.34% more requests correctly when compared to the original British Airways ASP algorithm. As British Airways has a large
Chapter 8 Assessment 140
amount of its revenue dependent on the judgement of it’s cargo yield analysts, a system that can aid the decision making process with this degree of accuracy (when compared to the existing ASP system) would greatly increase the confidence that a correct decision has been made. Even a small increase in performance would be extremely valuable to British Airways.
Percentage Correct on Test Set British Airways ASP 62.5%
Improved ASP 67.5%
Hybrid Solution 83.84%
Table 8.1 — Comparison of results for the original ASP algorithm, the improved implementation and the final hybrid solution.
Also another important improvement would be the saving in time of processing a request. Currently it takes approximately 1 hour (from the British Airways data) from when the request is received to when the final decision is made. This could be made almost real-time, where the analyst could say almost immediately if the request would be accepted or not. This would not only improve the service offered to the client, but also increase the volume of requests that can be processed by the analyst. This intum would help to increase the revenue of the company.
The use of such a hybrid system for cargo consignment, would give the following advantages:
1. Adaptability in the face of a changing environment, where the neural networks can be retrained as the cargo trends for a route or flight change.
2. Automation of all but the key decisions.
3. Combination of processing techniques to help make better quality decisions. 4. The cargo consignment process becomes profit and service driven rather than
capacity driven.
The British Airways cargo consignm ent application has show n the suitability o f the
Chapter 8________________________________Assessment_____________________________________ 141
8.5. Summary
The main goals of this research work have been to establish the theoretical integration issues of hybrid systems and to construct the O r b e r o n environment for applying hybrid systems. The main areas covered during this research have been the definition of a hybrid classification scheme, the design and implementation of an object- oriented hybrid environment (ORBERON) and the implementation of a real-world cargo consignment application from British Airways.
The hybrid classification scheme has been compared to a scheme devised by Medsker and Bailey [68] and proved to be simpler, did not contain redundant classes and therefore was easier to apply to various types of hybrid systems.
The object-oriented approach to hybrid systems and the O r b e r o n environment were compared with emerging integration technologies (such as Microsoft’s OLE and the Object Management Group’s Object Request Broker) that use object-oriented methods for communication and co-ordination of components. The functional aspects of components of the O r b e r o n environment are compared with the components found in these new integration techniques. The HANSA Framework and the ORBERON environment, are also compared with respect to functionality.
Finally, the results of the hybrid British Airways cargo consignment solution has been assessed with respect to the results produced by the currently used expert system implementation of the Acceptable Shipment Package algorithm. The results from the hybrid solution showed a marked improvement in the quality of decisions and performance of the cargo acceptance procedure.