Umeå University's logo

Classification networks, consisting of preprocessing layers combined with well-known classification networks, are well suited for medical data analysis. Additionally, by adjusting network complexity to corresponding complexity of data, the parameters in the preprocessing network can, in comparison with networks of higher complexity, be more precisely understood and also effectively utilised as decision limits. Further, a multivariate approach to preprocessing is shown in many cases to increase correctness rates in classification tasks. Handling network complexity in this way thus leads to efficient parameter estimations as well as useful parameter interpretations.

Many students feel insecure making their first attempts to solve programming problems. Despite finishing the introductory programming course successfully, these students refrain from pursuing their CS studies. Hence, this aversion towards problem solving and programming is not fully explained by lack of subject understanding and performance. In order to better understand the components of students’ comfort, a first attempt to model a student’s confidence regarding problem solving and programming has been made. The model consists of two dimensions; Course topic and Student’s mindset. Two questionnaires have been developed in order to capture if and how students’ confidence is affected by taking the CS1 course. Data has been collected for four course offerings with three different study programmes. Results confirm the suspicion that the confidence is lowered by the course, and that student groups with different ambition and motivation for taking the course seem to be affected by different aspects of the course.

Not being part of a larger SI-organisation has both advantages and disadvantages. In this paper we try to illustrate the advantages of doing SI small scale. In a large scale SI-organisation the supervisors are often not teachers themselves and/or not familiar with the practices of a specific course. To have teaching staff supervising a SIproject completely focused on one course is favourable in many ways. The decision to introduce SI was taken by the department of Computing Science to support the students at the introductory course in object oriented programming. This course demands a high level of abstract thinking and is very heavy on many of the new students majoring in computing science. In 2002 we started off with a general training course for eight new SI-leaders, but soon discovered that much could be gained from making the training specifically working with the course at hand.

Working with the course material in the training course makes it possible to use all ideas, experiences, and material developed during the training directly in the SI-meetings. The SI-leaders can prepare their information to the students; they can even start planning their first meeting. Another import part of the training is simulated SI-meetings. We “stage” them to illustrate different aspects of group dynamics and to control that not all problematic situations happen at one single meeting. Supervisor meetings are another important component of a SI-project. They benefit from small scale since they provide an excellent possibility for feedback and exchange of ideas if the supervisors themselves have experience in teaching the actual course. Working in small scale with one specific course also makes it possible to refine the study strategies and techniques used in the SI-meetings. The course is heavily targeted towards problem solving and this has influenced the meetings in different ways. For example it is not uncommon to let some of the students try out ideas in the computer labs and report back to the SI-group during a meeting. Results from six projects finished since 2002 will be presented and discussed. We have seen that the performance and grades are higher in the group attending SI and we are currently doing follow ups to check the retention rate.

In this paper, several ways to handle missing data, e.g. removing cases, mean imputation, and multiple imputation, are described and discussed. The Pima-Indians-Diabetes data set is used as a case study. This particular data set is interesting to use since it has not been obvious to all users that it actually contains a substantial amount of missing data. The data set is described in detail and the methods for coping with missing data mentioned in the text is applied on the data set.

The preprocessing perceptron is used to train decision support systems on the data sets. A sketch of a way to impute missing data using the preprocessing perceptron is also proposed and discussed. The accuracy of the trained decision support systems, at the optimal efficiency point, lied in the interval 76-82% for the different methods. The highest values were obtained when all missing data cases were removed both from the test and the training set. This is, however, not a good way to handle missing data since the resulting decision support system is biased. Furthermore it will not be able to handle missing data when used on real data in the future. The results of the remaining methods were surprisingly similar, a reason for this might be that the data set used is rather large. Differences between methods would probably be larger in a smaller data set with larger amount of missing data.

Reliable results are crucial when working with medical decision support systems. A decision support system should be reliable but also be interpretable, i.e. able to show how it has inferred its conclusions. In this thesis, the preprocessing perceptron is presented as a simple but effective and efficient analysis method to consider when creating medical decision support systems. The preprocessing perceptron has the simplicity of a perceptron combined with a performance comparable to the multi-layer perceptron.

The research in this thesis has been conducted within the fields of medical informatics and intelligent computing. The original idea of the production line as a tool for a domain expert to extract information, build decision support systems and integrate them in the existing system is described. In the introductory part of the thesis, an introduction to feed-forward neural networks and fuzzy logic is given as a background to work with the preprocessing perceptron. Input to a decision support system is crucial and it is described how to gather a data set, decide how many and what kind of inputs to use. Outliers, errors and missing data are covered as well as normalising of the input. Training is done in a backpropagation-like manner where the division of the data set into a training and a test set can be done in several different ways just as the training itself can have variations. Three major groups of methods to estimate the discriminance effect of the preprocessing perceptron are described and a discussion of the trade-off between complexity and approximation strength are included.

Five papers are presented in this thesis. Case studies are shown where the preprocessing perceptron is compared to multi-layer perceptrons, statistical approaches and other mathematical models. The model is extended to a generalised preprocessing perceptron and the performance of this new model is compared to the traditional feed-forward neural networks. Results concerning the preprocessing layer and its connection to multivariate decision limits are included. The well-known ROC curve is described and introduced fully into the field of computer science as well as the improved curve, the QROC curve. Finally a tutorial to the program trainGPP is presented. It describes how to work with the preprocessing perceptron from the moment when a data file is provided to the moment when a new decision support system is built.

In recent years, students have become less active, resulting in lower attendance in lectures and practical sessions. In addition to this, the number of students enlisting in our programmes has decreased. Moreover the passing rates for initial courses have dropped severely. This generates problems because the students failing first year courses cannot move on to higher level courses. Not only can this be devastating for individual students, but it can also affect the variety of higher level courses. In an attempt to prevent these problems we focused on the introductory programming courses (CS1) in order to enhance the opportunities for the students to become successful. One action taken was a research project initiated to radically change the way object-oriented programming is taught in CS1. Another action was to introduce the Supplemental Instruction programme (SI). SI helps students master content while they develop and integrate learning and study strategies. This paper will give a short introduction to these actions. Results are presented along with a discussion concerning the problems in teaching object-oriented concepts and problem solving.