Reliability Improvement On Feasibility Study For Selection Of Infrastructure Projects Using Data Mining And Machine Learning

With the progressive development of infrastructure construction, conventional analytical methods such as correlation index, quantifying factors, and peer review are no longer satisfactory in support for decision-making of implementing an infrastructure project in the age of big data. This study proposes using a mathematical model named Fuzzy-Neural Comprehensive Evaluation Model (FNCEM) to improve the reliability of the feasibility study of infrastructure projects by using data mining and machine learning. Specifically, the data collection on time-series data, including traffic videos (278 Gigabytes) and historical weather data, uses transportation cameras and online searching, respectively. Meanwhile, the researcher sent out a questionnaire for the collection of the public opinions upon the influencing factors that an infrastructure project may have. Then, this model implements the backpropagation Artificial Neural Network (BP-ANN) algorithm to simulate traffic flows and generate outputs as partial quantitative references for evaluation. The traffic simulation outputs used as partial inputs to the Analytic Hierarchy Process (AHP) based Fuzzy logic module of the system for the determination of the minimum traffic flows that a construction scheme in corresponding feasibility study should meet. This study bases on a real scenario of constructing a railway-crossing facility in a college town. The research results indicated that BP-ANN was well applied to simulate 15-minute small-scale pedestrian and vehicle flow with minimum overall logarithmic mean squared errors (Log-MSE) of 3.80 and 5.09, respectively. Also, AHP-based Fuzzy evaluation significantly decreased the evaluation subjectivity of selecting construction schemes by 62.5%. It concluded that the FNCEM model has strong potentials of enriching the methodology of conducting a feasibility study of the infrastructure project