← CSC 316 Research Methodology in Computing
Research Design and Methods in Computing
Notes
Choosing a Research Design
Computing research employs diverse methodologies depending on the underlying research questions. The three primary paradigms are theoretical research, empirical research, and design science research. Theoretical research relies on mathematical proofs and formal logic to analyze complexity, correctness, and automata behavior. Empirical research measures existing computing systems or human-computer interactions through quantitative benchmarks, telemetry logs, or user experiments. Design Science Research focuses on the creation and evaluation of innovative IT artifacts.
Design Science Research Methodology (DSRM)
DSRM is the dominant methodology in applied computing. It follows a structured process: problem identification, objective definition, artifact design and development, demonstration, evaluation, and scientific communication. The artifact produced may be an algorithm, system architecture, human-computer interface, or formal notation. The evaluation phase must be rigorous, using metrics such as memory overhead, CPU cycle utilization, convergence speed, precision, and recall.
Quantitative versus Qualitative Approaches in Computing
Quantitative methods dominate systems and algorithm research where hypotheses are accepted or rejected based on numeric evidence, statistical variance, and regression models. Conversely, qualitative methods are widely used in software engineering process improvement and human-computer interaction (HCI). Qualitative techniques include thematic analysis of developer interview transcripts, cognitive walkthroughs, and observational user experience studies.
Benchmarking and Experimental Rigor
To ensure scientific reproducibility, computing experiments require strict environment controls. Researchers must document hardware configurations, compiler flags, operating system versions, and hyperparameter tunings. Benchmarking demands representative workloads, standardized baseline algorithms, multiple execution runs to eliminate transient OS noise, and reporting of confidence intervals.
Practical example
A computing student conducts research on crop disease detection for smallholder maize farmers in Trans-Nzoia County who use low-cost, resource-constrained Android smartphones.
Methodology Selected: Design Science Research Methodology (DSRM).
Artifact: A quantized, lightweight convolutional neural network (CNN) compressed via post-training 8-bit quantization to run locally without internet connectivity.
Experimental Setup: The student curates a dataset of 4,000 field images of Northern Corn Leaf Blight taken under direct sunlight in Kitale. The benchmark baseline is MobileNetV3.
Evaluation Metric: The compressed artifact is evaluated on physical low-tier devices (2GB RAM, MediaTek processor). The researcher records inference speed (milliseconds), RAM usage (megabytes), battery drain (milliampere-hours per 100 inferences), and Top-1 classification accuracy across 50 independent test runs.
Quiz
1. Why are multiple experimental execution runs necessary when evaluating algorithmic performance?
2. In Design Science Research Methodology (DSRM), what constitutes an acceptable IT artifact?
3. In the Trans-Nzoia maize disease detection study, why was post-training quantization applied?
Sign in to submit and securely grade your quiz.