Predicting project duration using a coupled artificial neural network and Taguchi method approach

Abstract

Accurate project duration prediction is increasingly important for management because it defines the expected timeline for project realization. This study utilizes an integrated approach combining neural networks with the Taguchi method to forecast the time required to complete projects within predetermined deadlines. The methodology involves modelling and simulating the network of project activities to estimate the total average project duration. ration. The neural network model uses input variables such as success probability, effort, and learning factor to predict the total time necessary for project completion. The total average project duration is the output variable that is critical during the design phase. Subsequently, the Taguchi method optimizes the neural network’s outputs, incorporating mean squared error (MSE) values to enhance predictive accuracy. accuracy. This study underscores the efficacy of artificial neural networks (ANNs) as predictive tools, demonstrating their ability to meticulously estimate project duration. The proposed method’s efficiency and applicability are demonstrated by MATLAB simulation analyses, highlighting its effectiveness in precise deadline estimation. In the realm of engineering, ANNs stand tall as formidable predictive tools, their efficacy underscored by this study’s successful application. By harnessing ANNs and simulation data, this research crafts a predictive model adept at estimating the average total duration of projects. Through meticulous consideration of crucial parameters like the probability of success and effort factors, the model emerges as a beacon of accuracy within the design domain. Future research will explore the effects of additional parameters on activity networks and alternative transfer functions, as well as the potential integration of reinforcement algorithms to improve resource allocation, risk management, and project outcomes through online training data.