Estimation of Effort in Software Cost Analysis for Heterogenous Dataset using Fuzzy Analogy

One of the significant objectives of software engineering community is to use effective and useful models for precise calculation of effort in software cost estimation. The existing techniques cannot handle the dataset having categorical variables efficiently including the commonly used analogy method. Also, the project attributes of cost estimation are measured in terms of linguistic values whose imprecision leads to confusion and ambiguity while explaining the process. There are no definite set of models which can efficiently handle the dataset having categorical variables and endure the major hindrances such as imprecision and uncertainty without taking the classical intervals and numeric value approaches. In this paper, a new approach based on fuzzy logic, linguistic quantifiers and analogy based reasoning is proposed to enhance the performance of the effort estimation in software projects dealing with numerical and categorical data. The performance of this proposed method illustrates that there is a realistic validation of the results while using historical heterogeneous dataset. The results were analyzed using the Mean Magnitude Relative Error (MMRE) and indicates that the proposed method can produce more explicable results than the methods which are in vogue.

💡 Research Summary

Software effort estimation remains a pivotal activity in project planning, budgeting, and risk management. Traditional models such as COCOMO, Function Point analysis, and a variety of regression‑based machine‑learning techniques have achieved reasonable accuracy when the input variables are purely numerical. However, real‑world project datasets are often heterogeneous: they contain a mixture of numeric attributes (e.g., size in KLOC, duration) and categorical or linguistic attributes (e.g., “high”, “medium”, “low” complexity, team experience, requirement clarity). Existing analogy‑based estimation methods typically treat categorical data by one‑hot encoding or simple binary similarity, which discards the semantic distance between linguistic terms and fails to capture the inherent fuzziness of human‑generated descriptors.

The paper titled “Estimation of Effort in Software Cost Analysis for Heterogenous Dataset using Fuzzy Analogy” proposes a novel framework that integrates fuzzy set theory with analogy‑based reasoning to overcome these shortcomings. The core contributions can be summarized as follows:

1. Fuzzy Representation of Linguistic Variables – Each linguistic grade is modeled by a fuzzy membership function (triangular, Gaussian, or trapezoidal). For example, the term “high” is expressed as μ_high(x) = max(0, 1‑|x‑c|/w), where c is the central value and w the spread. This continuous representation allows a project attribute value to belong partially to multiple linguistic categories, preserving the nuance lost in crisp encoding.

2. Use of Fuzzy Quantifiers – Global statements such as “most projects have low complexity” are quantified using fuzzy quantifiers (e.g., “most” ≈ 0.7‑0.9). These quantifiers are incorporated into the similarity calculation, providing a dynamic weighting scheme that reflects the overall distribution of categorical attributes across the historical dataset.

3. Fuzzy Distance Metric for Analogy Selection – The similarity between a target project i and a historical project j is computed as
   D(i,j) = √