Overview
A methodology provides an organized approach to conducting a data science investigation. It helps practitioners move from an initial question to a defensible result by establishing appropriate activities, decision points, analytical standards, and evaluation procedures. Different projects may require different methodologies because scientific research, business analytics, predictive modelling, and large-scale machine learning do not always have identical objectives or constraints.
Methodology Vs Method Vs Technique
Methodology
The overall framework of principles and organized procedures that guides an investigation. It explains how and why different activities are arranged.
Method
A particular approach used to perform an activity, such as a survey, statistical analysis, cross-validation, or experimental comparison.
Technique
A specific procedure or implementation used within a method, such as k-means clustering, decision-tree splitting, or mean imputation.
Algorithm
A precise computational procedure consisting of defined steps for solving a problem or performing a calculation.
Example
A project may follow CRISP-DM as its methodology, use supervised learning as its analytical method, apply a decision tree as a modelling technique, and implement the model using a particular algorithm.
Major Methodological Frameworks
| Name | Full Form | Overview | Phases | Strengths | Limitations | Major Activities | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
CRISP-DM | Cross-Industry Standard Process for Data Mining | CRISP-DM is a widely recognized framework for organizing data mining and data science projects. It provides six connected phases and emphasizes understanding the project objective before selecting technical solutions. |
|
|
| |||||||||||||||
KDD | Knowledge Discovery in Databases | KDD describes the broader process of discovering useful, valid, novel, and understandable knowledge from data. Data mining is one important stage within this process rather than a synonym for the entire KDD process. |
|
|
| |||||||||||||||
Team Data Science Process | The Team Data Science Process is a collaborative framework for developing predictive analytics solutions through coordinated team activities, shared artifacts, project planning, and technical implementation. |
|
|
|
Core Methodological Principles
| Principle | Explanation |
|---|---|
Problem-first investigation | Begin with a meaningful question and determine what evidence is required before choosing an algorithm. The method must serve the objective rather than the objective being artificially selected to fit a preferred technique. |
Evidence-based analysis | Analytical conclusions should be supported by relevant observations, valid measurements, appropriate statistical reasoning, and transparent evaluation. |
Data quality awareness | Recognize that missing values, measurement errors, sampling bias, inconsistent definitions, and unreliable sources can influence conclusions. |
Appropriate method selection | Select methods according to the research question, data structure, assumptions, interpretability requirements, computational resources, and consequences of incorrect results. |
Reproducibility and transparency | Document data sources, transformations, software versions, parameters, analytical decisions, and evaluation procedures so that the work can be examined and reproduced. |
Iterative improvement | Treat unexpected results and evaluation failures as evidence that may require revisiting earlier assumptions, data, or analytical choices. |
Ethical and responsible analysis | Consider privacy, consent, fairness, potential harm, data ownership, and the appropriate use of analytical outputs throughout the project. |
Methodology Selection Factors
| Factor | Explanation |
|---|---|
Research objective | Descriptive, predictive, explanatory, causal, and prescriptive questions require different analytical approaches. |
Data characteristics | Data volume, structure, measurement quality, variable types, temporal properties, and sampling procedures influence method selection. |
Domain requirements | Scientific and professional domains may impose specific measurement standards, regulatory obligations, or interpretability requirements. |
Available resources | Time, budget, computing hardware, software, expertise, and data access can determine which approaches are feasible. |
Evaluation requirements | The project must establish how success will be measured and what level of uncertainty or error is acceptable. |
Deployment context | A method intended for real-time prediction may require different computational and operational properties from one used for offline scientific analysis. |
Methodological Approaches
| Approach | Explanation |
|---|---|
Exploratory approach | Investigates data to discover patterns, identify unusual observations, generate hypotheses, and develop an understanding of the problem without beginning with a narrowly specified predictive model. |
Confirmatory approach | Tests predefined hypotheses or relationships using appropriate statistical procedures and a clearly specified analytical design. |
Predictive approach | Develops models that estimate unknown or future outcomes from available information and evaluates their performance on data not used for model fitting. |
Causal approach | Investigates whether a factor produces a change in an outcome using suitable experimental or observational causal-inference designs. Ordinary correlation alone is insufficient to establish causality. |
Prescriptive approach | Uses predictions, constraints, objectives, and optimization methods to recommend actions or decisions. |
Worked Example
Problem
A university wants to investigate factors associated with students missing scheduled classes and develop an evidence-based attendance support program.
Methodological Plan
Define the objective and distinguish between understanding attendance patterns and predicting future absence.
Identify suitable data sources, such as attendance records, course schedules, and voluntarily collected student feedback.
Assess data quality, representativeness, privacy requirements, and possible missing information.
Explore attendance patterns across courses, days, and relevant time periods.
Select an appropriate analytical approach based on the research question.
If prediction is required, construct and validate a model using suitable training and evaluation procedures.
Examine whether the results differ across relevant groups and whether any observed relationships can reasonably be interpreted.
Present findings with limitations and recommendations.
Monitor whether an approved support program produces the intended outcomes.
Interpretation
A suitable methodology prevents the project from treating prediction as the only goal. It also ensures that data collection, statistical interpretation, privacy, and evaluation are considered before recommendations are made.