Вы находитесь на странице: 1из 6

Selection of Lawn and Garden Tractors Using an Expert System

Babak Negahban, Richard C. Fluck, W. David Shoup


MEMBER ASAE MECHANIZATION MEMBER ASAE

ABSTRACT O known guidelines have existed for consumer selection of lawn and garden tractors. Certain rules for selection were developed from a knowledge extraction process from experts (top salespersons) for two major brands of outdoor products. Their knowledge was used in constructing expert systems to assist consumer selection. Each brand required its own unique programming logic in order to accomplish proper selection. The accuracy of the expert systems was verified by comparing the model output to salesmens' selections.

PROBLEM STATEMENT Selecting the appropriate tractor for one's needs has serious consequences. A lawn and garden tractor is a long-lived and expensive piece of equipment. A tractor that is too small will require excessive time to operate and have short life and poor reliability, whereas a tractor that is too large will have needlessly high fixed costs. In addition, options are important for they enable certain operations which are otherwise excluded. Thousands of lawn and garden tractors are purchased annually with little knowledge of tractor operation, leading to mistakes in selection. Guidelines did not exist in the lawn and garden tractor industry for assisting customers and beginning salespersons in selecting lawn and garden tractors. Most often the consumer must depend on the judgment of the garden tractor salesperson to recommend the best garden tractor that meets the buyer's needs. OBJECTIVE The objective of this work was to develop an expert system for lawn and garden tractor selection. This system could be used in aiding salesmen and consumers in selecting appropriate equipment. REVIEW OF LITERATURE American Society of Agricultural Engineers Standard S323.2 (ASAE, 1986) defines a lawn and garden tractor
Article was submitted for publication in August, 1987; reviewed and approved for publication by the Power and Machinery Division of ASAE in March, 1988. Approved for publication as Florida Agricultural Experiment Station Journal Series No. 8326. The use of trade names in this publication does not imply endorsement by the University of Florida of the products named nor criticism of similar ones not named. The authors are: BABAK NEGAHBAN, formerly Graduate Research Assistant, University of Florida and currently Teaching Assistant, University Center, Dschang, Cameroon; RICHARD C. FLUCK, Professor; and W. DAVID SHOUP, Associate Professor, Agricultural Engineering Dept., University of Florida, Gainesville. Vol. 4(3):September, 1988

as a self-propelled machine that is designed to supply power for home lawn, home garden and yard maintenance implements. The tractor should have some means to lift implements, and implements should be separate from the tractor. No literature was found that provided any guidelines for lawn and garden tractor selection. Very little research has been published concerning lawn and garden tractors despite Outdoor Power Equipment Institute's (1986) forecast of 1,020,000 units of riding lawn and garden tractors to be shipped in the U.S. in 1986, or about two billion dollars of product. Past shipments indicate there are a total of several million in use. By comparison, the agricultural tractor market in the U.S. was only 108,795 units in 1985, or approximately 4.7 billion dollars of product (Implement & Tractor, 1987). PROCEDURES Identifying and Interviewing the Experts Florida regional managers for Ford, John Deere and Kubota were identified and contacted. Each regional manager was asked to name three of the top salespersons in the North Central Florida area. The criteria for selecting the salespersons were their records of high sales volume and long-term sales experience. Top salespersons from Kubota and John Deere organizations were chosen to participate in a knowledge extraction process regarding how they selected a tractor to fit a customer's needs. Selection rules were derived from the extraction process on which to base the expert system. The salespersons identified by the regional managers were contacted by phone and interview dates were obtained. The Sondeo approach was used in the interview process (Hildebrand, 1980). In the Sondeo approach, interviews are done in a very casual and informal manner. Very few questions are preplanned. The salespersons were asked to express in their own words how they go about recommending a garden tractor to a potential customer. Selecting a Decision Making Tool Weiss and Kulikowski (1984) defined an expert system as a computer model that can handle real world problems requiring an expert's interpretation. Helping a person select a proper garden tractor certainly does require assistance from an expert. Expert systems mimic the deductive reasoning of human experts (Negoita, 1985). In order to develop a source of information to aid customers, the need for interaction is essential. Knowledge systems or artificial intelligence systems are necessary for true, logically oriented problem solving and for two way communication. Expert systems can
211

1988 American Society of Agricultural Engineers 0883-8542/88/0403-0211502.00

facilitate an analysis of each consumer's unique and specific needs. Expert systems can show the consumer lines of reasoning, can show the pool of information being used, can work with nonquantitative data and can work with incomplete information; all conditions normally found in selling. Thus this new tool was tried. Developing the Expert System Once the expert knowledge had been gathered, the next step was organization of the knowledge into a usable format. Information that the expert salespersons had provided was divided into four different categories. The first category was information critical to the selection of garden tractors. The expert system was unable to make a recommendation if this information was not available. Relationships between size of work area, type of activities and recommended garden tractor model were considered to be essential. The second category was information that did not influence the garden tractor selection but did influence the tone of the recommendation. The customer's evaluation of his previous garden tractor, if any, was the main component of this category. For example, if the customer felt that his previous garden tractor was too small and the expert system was recommending an even smaller garden tractor, the tone of the recommendation would be more cautious and would ask the customer to look further into his special conditions. On the other hand, if both the customer and the expert system agreed on the incapability of the previous garden tractor, the tone of the recommendation would be much stronger. The third category consisted of information that had no bearing on the recommended garden tractor but nevertheless was appropriate to be included in the expert system output, such as information about garden tractor maintenance and safety recommendations. The last category consisted of information obtained in the interviews with the garden tractor salespersons that was irrelevant to the garden tractor selection process. Such information was ignored. The heuristic information obtained from the interviews of salespersons of two brands of tractors was sufficiently different so that a single generic expert system model could not be constructed to include both. Therefore, separate models were constructed for the two brands. The first two above categories of information were organized into two decision trees for garden tractor selection. Creating the Source Code for the Expert System Insight 2 + (Level Five Research, 1985) was the expert system shell selected to incorporate the decision rules and form the model. Reflex (Borland, 1985) was used to enhance the input and output of Insight 2 + . This enabled the expert system to be "user friendly" and display results in a graphical format. These software programs were selected on the basis of availability, cost and ease of use. The decision trees for the models included approximately 4500 steps for the John Deere model and 3500 steps for the Kubota model. The decision tree for the selection of Kubota lawn and garden tractors is shown in Fig. 1. Once the knowledge base had been transferred to workable decision trees, the next step
212

was to develop the source code for Insight 2 + programs that would duplicate the information in the decision trees. Most of the body of programs created with Insight 2 + is contained in the goals and rules sections. The goals for the garden tractor decision selection system were relatively simple. The intention was for the system to be able to recommend a garden tractor model for the specific needs of a particular customer. Therefore, each garden tractor model that was covered in the knowledge base was defined to be a goal in the program. An example of one of the possible goals in the program is: 1. Appropriate Garden Tractor is KUBOTA G3200
HAVE YOU PREVIOUSLY OWNED A GARDEN TRACTOR '

WAS IT A KUBOTA GARDEN TRACTOR ?

WHAT MODEL WAS THE OLD GARDEN TRACTOR ?

WHAT WAS THE OLD GARDEN TRACTOR ENGINE'S HORSEPOWER?

DO YOU FEEL THAT YOUR OLD GARDEN TRACTOR WAS TOO SMALL, JUST RIGHT OR TOO LARGE ?

DO YOU PLAN TO DO GARGEtt&iG WfXH YOUR GARDEN TRACTOR t

WHAT IS YOUR ESTIMATED MOWING AREA (ACRES) ?

20 OR MORE 10 TO 15 "1 SELECT B 7200 SELECT G 5200 GARDEN TRACTOR NOT APPROPRIATE

5 TO IP' SELECT G 6200

SELECT B 8200

0 TO 1.5

| [ 1.5 TO 2 . 5 ~ | SELECT G 4200

SELECT G 3200

WHAT t $ YGU& ESTIMATED MQW-IN6 AREA NACRES* 1

(Fig. 1continues) APPLIED ENGINEERING in AGRICULTURE

2 OR LESS 1 SELECT B 5200

2 TO 6 " |

6 TO I 2 ~ |

| 12 TO 2o""|

1 2 0 OR MORE GARDEN TRACTOR NOT APPROPRIATE]

SELECT 8 6200

SELECT B 7200

SELECT B 8200

3 5 TO 10

WHAT IS YOUR ESTIMATED GARDENING AREA (ACRES)?

:o
SELECT B 6200 SELECT B 7200

SELECT B 8200

GARDEN TRACTOR NOT APPROPRIATE!

H'O TO 15

WHAT IS YOUR ESTIMATED GARDENING AREA (ACRES)?

SELECT B 7200

SELECT B 8200

GARDEN TRACTOR NOT APPROPRIATE

0
WHAT IS YOUR ESTIMATED GARDENING AREA (ACRES)?

SELECT B 8200

GARDEN TRACTOR NOT APPROPRIATE^

Fig. 1Decision tree for the selection of Kubota lawn and garden tractors. Vol. 4(3):September, 1988

213

Insight 2 + would ask the user the relevant questions and try to verify this goal. If what the user entered would not verify this goal, then Insight 2 + would try to verify another goal. Through the process of elimination, Insight 2 + would eventually find a goal that it could verify and recommend a specific garden tractor model. To take care of situations where no garden tractor is recommended one final goal was added to the list. This goal stated that a garden tractor is not recommended for the kind of work the customer wants to do. An example of such a situation was that the customer wanted to do gardening on more than 20 acres. The next step in the program was to define the rules needed to achieve the stated goals. It must be understood that there must be a rule defined for every possible pathway in the decision tree that has been created. If a possible pathway in the decision tree is left out of the program, Insight 2 + will fail to make a recommendation when the customer's situation is the same as the missing pathway. In the decision tree for the garden tractor selection there are from 16 to 48 different situations in which any single garden tractor model is recommended. Therefore, for any given tractor model there are from 16 to 48 different rules that can achieve the goal of recommending that model. For example, rule number 10 for recommending the Kubota B-7200 tractor is as follows: RULE For Kubota Option 10 If You have previously owned a garden tractor AND It was a KUBOTA garden tractor AND Your previous garden tractor model IS too large AND You will use this tractor for gardening AND Estimated mowing area in acres < 10 AND Estimated gardening area in acres > 6 AND Estimated gardening area in acres < = 12 THEN Appropriate Garden Tractor is KUBOTA B7200 AND Selection : = B7200 AND Mowing : = 15 AND Gardening : = 12 AND DISPLAY Recommendation 8 The first line in this rule is the rule title and a declaration that what comes after is a rule. The second through fourth lines determine the tone of the recommendation (this issue has been previously explained). The next four lines determine whether the goal (line 9) can be verified. The last four lines are used to make Insight 2 + more user friendly. The entire expert system for garden tractor selection was a collection of rules similar to the one presented above. Insight 2 + sorts through these rules until it can carry out one to the very end and verify its stated goal. One final consideration was to develop a program that would be easy to use, even for a person with no computer knowledge. This required that there be adequate instructions on the use of the computer. A number of screens were created at the beginning of the program that presented the user with detailed information about the computer keyboard and the keys that he would need to use (i.e. number keys, curser keys and function keys). In addition, every time the user was prompted with a
214

question he was given instructions on how to enter his answer. In order to make questions understandable, detailed explanations were presented to the user each time he was prompted with a question. In some cases the user had the option of going to a second screen of information for more detailed explanations. Finally, the user received a full screen of information that discusses the recommended garden tractor (the DISPLAY command line). This screen also included opinions about the customer's previous graden tractor, if any. If differences of opinion existed between the customer and the expert system, this was also included on this screen. Essentially, an attempt was made to create an expert system that could make effective recommendations about garden tractors and also be friendly enough to be easily used by the customer. The expert system developed provides each dealership with capability to accumulate customer data. The customer accesses the expert system via an IBM compatible microcomputer which can be located at the dealership or which can be a portable unit provided by a salesperson. The accumulated data could be used to improve or update the expert system. Reflex stores each response from each customer for any given time frame in a data base. Reflex also contains a report writer program which organizes the data into tables or graphs. Field Testing the Expert System Once the two models were created, each was taken for evaluation to eight expert salespersons for that brand in the region. Salespersons were given the expert system in order to test hypothetical consumer situations. The salespersons then evaluated the expert system by comparing their conclusions with those of the expert system. A questionnaire was used as a tool for evaluating the salesmen's agreement that the expert system properly reflected their decision process. The questionnaire consisted of 14 statements; in addition to approximation of reality, it also included statements relating to the model's ease of use and value to the salesperson. Salespersons were asked to give each statement a score between 1 and 5; 1 being excellent, 2 good, 3 average, 4 fair and 5 poor. Salespersons were also given the opportuntiy to write comments on the second page of the evaluation sheet. RESULTS AND DISCUSSION All of the evaluators stated verbally that they felt the expert system properly reflected the decision process for making recommendations. An analysis of the questionnaires is provided in Table 1. The mean scores for all of the statements were between 1.250 and 1.875. The range was from 1 to 5. The standard deviations of the scores ranged between 0.447 to 0.957. The 95% confidence intervals for the mean scores ranged from a minimum low of 0.880 to a maximum high of 2.334. Amount of instructions, ease of use after workshop and usefulness of saving customer information received the lowest (best) mean scores (1.250) in the evaluation. The first two factors emphasized the user friendliness of the expert system.
APPLIED ENGINEERING in AGRICULTURE

TABLE 1. STATISTICAL ANALYSIS OF MODEL EVALUATION SCORES Statement Reasonable amount of instructions Language is understandable Program is concise, not wordy Screen legibility Could be used after a short workshop Easy to use Approximates reality Program proceeds in a logical manner to solve problem Makes a reasonable recommendation Potential as a salesperson training tool Potential as a sales tool Saving customer information is useful Information summary and graphs created from retained information are useful Recommend the use of this program Mean score 1.250 1.313 1.625 1.875 1.250 1.312 1.625 1.375 1.313 1.438 1.500 1.250 Std. dev. 0.447 0.479 0.719 0.957 0.447 0.602 0.619 0.619 0.479 0.512 0.730 0.500 95% C.I. 1.03-1.47 1.09-1.55 1.27-1.98 1.41-2.34 1.03-1.47 1.07-1.61 1.32-1.93 1.07-1.68 1.08-1.55 1.18-1.69 1.14-1.86 0.88-1.37

recommending larger models in each company's product line. Kubota salespersons had criteria for their full line of lawn and garden tractor models. Consequently, the expert system for Kubota garden tractor selection can make solid recommendations on the use of the larger models in appropriate situations. On the other hand, John Deere salespersons did not have criteria for selecting their larger garden tractor models. Rather, they felt that many of their larger models were selected on the basis of luxury features rather than capacity. As a result, the expert system for John Deere garden tractor selection does not make recommendations for the larger models. John Deere salespersons could not provide sufficient information for a logic tree to support the expert system's making decisions for their larger models. Limitations of the Models The models have limitations. The most obvious limitations are that they cover only garden tractors and only those of the John Deere and Kubota brands. The expert system technique could be expanded into other agricultural machinery and manufacturers. Another likely limitation is the models' geographic limitations. Since the expertise of the salespersons interviewed is only in North and Central Florida, the models can only be certain to handle situations that are common to that area. To expand the model into other geographic areas would require interviewing garden tractor experts in each homogeneous geographic zone and incorporating their knowledge into the model. CONCLUSIONS An evaluation of the information obtained in this study has resulted in the following conclusions: 1. While manufacturers of lawn and garden tractors provide limited guidelines for the selection of lawn and garden tractors, no decision criteria are provided except for salespersons' guidance. 2. The most important factors involving lawn and garden tractor selection are size of work area and type of activity. 3. Differences between brands in the obtained information and tractor models required that two distinct expert systems rather than a generic expert system be constructed for the two lines. 4. The minimum threshold land area for which machines are recommended is larger with the Kubota model than with the John Deere model. Further, the maximum land area for which machines are recommended is larger with the Kubota model than with the John Deere model. 5. It is possible to verify an expert system by field study. 6. Expert systems can be developed for non-technical areas such as marketing. While expert systems could be used in marketing agricultural equipment, these applications would require entirely different logic trees. References
1. ASAE. 1986. ASAE Standards 1986. ASAE, St. Joseph, MI 49085. 2. Borland. 1985. Reflex-The Analyst. Borland/Analytica Inc., 4585 Scotts Valley Drive, Scotts Valley, CA. 3. Hildebrand, P. E. 1980. Combining disciplines in rapid

1.375 1.500

0.619 0.632

1.07-1.68 1.19-1.81

The range of scores was from 1 (best) to 5 (worst).

Most evaluators felt that accumulating customer data was very important. However, evaluators at some of the smaller dealers felt that this feature would not be of much use to them. They felt that they had such a small number of customers that they could mentally keep track of their customer base. Nevertheless, they gave high scores to the statement concerning this feature because they understood how such an option could be useful in other situations. The larger dealerships, on the other hand, liked this feature very much. One dealer felt that if he had had such a capability during the previous year, he would not have been stuck with some of the models that he had when interviewed. A number of differences exist between the way the Kubota and John Deere salespersons perceive the abilities of each of their garden tractor models. As a result, there are a number of differences between the expert systems developed for each company's product line, because the "experts" do not always agree. Therefore the two expert systems did not always recommend similar powered models in a given situation. John Deere salespersons recommended their lawn and garden tractors for use on areas up to 5 acres. On areas larger than 5 acres they recommended a larger line of their tractors termed utility tractors. In contrast, Kubota salespersons recommended their lawn and garden tractors for use on areas as large as 20 acres. As a result, the expert system for Kubota lawn and garden tractors can recommend lawn and garden models for a broader range of customer situations. One of the major differences between the expert systems for the two companies is the criteria used for
Vol. 4(3):September, 1988

215

appraisal: The Sondeo approach. Agr. Adm. 8:423-432. 4. Implement & Tractor. 1987. 50th market statistics issue. 102(11):7,14. 5. Level Five Research, Inc. 1985. Insight 2 + . Level Five Research, Inc., Indialantic, FL. 6. Negoita, C. V. 1985. Expert systems and fuzzy systems. The

Benjamin/Cummings Publishing Company, Inc., Menlow Park, CA. 7. Outdoor Power Equipment Institute. 1986. Industry Outlook. Bolens Corporation, Port Washington, WI. 8. Weiss, S. M., and Kulikowski, C. A. 1984. A practical guide to designing expert systems. Rowman & AUanheld Publishers, Totowa, NJ.

216

APPLIED ENGINEERING in AGRICULTURE

Вам также может понравиться