Analysing Variance in Elevator Installation Costs
Saamin, Maryam (2025)
2025
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-202505038818
https://urn.fi/URN:NBN:fi:amk-202505038818
Tiivistelmä
This research, drawing from historical data, investigates discrepancies between estimated and actual elevator installation costs within KONE projects. The core objective is to pinpoint both internal factors, such as resource availability and project management processes, and external factors, including site conditions and subcontracting issues, that significantly contribute to these cost variations. By identifying these key factors, the research seeks to develop predictive models that can enhance the accuracy of cost estimations.
To achieve this, the research employs both regression and classification analyses. While regression models were used to analyse feature importance, the research indicated that classification models, especially Random Forest, were more effective in categorizing cost discrepancies, specifically overruns and underruns. This suggests that using categorical distinctions is better at capturing underlying data patterns than relying on continuous predictions. To predict a specific cost category, such as Actual Fitter Cost, the Gradient Boosting Regressor demonstrated a slightly better fit compared to Random Forest and Linear Regression. An Artificial Neural Network (ANN) model was also applied to capture complex patterns in the data, strengthening the predictive accuracy. Furthermore, mathematical approaches such as the Delta P approach and the Root Sum of Squares (RSS) method were employed to analyse cost contributions.
The Delta P approach highlighted subcontract costs as the most significant contributor to discrepancies, while the Root Sum of Squares (RSS) method identified fitter costs as having the most substantial impact. The study underscores the significant impact of subcontract and fitter costs on project expenses, providing insights for improved decision-making, resource management, and cost reduction. Ultimately, the research aims to improve forecasting accuracy and manage cost variations, leading to more consistent project delivery, better customer trust, and reduced unexpected costs for KONE.
To achieve this, the research employs both regression and classification analyses. While regression models were used to analyse feature importance, the research indicated that classification models, especially Random Forest, were more effective in categorizing cost discrepancies, specifically overruns and underruns. This suggests that using categorical distinctions is better at capturing underlying data patterns than relying on continuous predictions. To predict a specific cost category, such as Actual Fitter Cost, the Gradient Boosting Regressor demonstrated a slightly better fit compared to Random Forest and Linear Regression. An Artificial Neural Network (ANN) model was also applied to capture complex patterns in the data, strengthening the predictive accuracy. Furthermore, mathematical approaches such as the Delta P approach and the Root Sum of Squares (RSS) method were employed to analyse cost contributions.
The Delta P approach highlighted subcontract costs as the most significant contributor to discrepancies, while the Root Sum of Squares (RSS) method identified fitter costs as having the most substantial impact. The study underscores the significant impact of subcontract and fitter costs on project expenses, providing insights for improved decision-making, resource management, and cost reduction. Ultimately, the research aims to improve forecasting accuracy and manage cost variations, leading to more consistent project delivery, better customer trust, and reduced unexpected costs for KONE.