Machine learning-based B2C software project success prediction model in Indonesia

The success of a software project is a crucial factor in the information technology industry, but it is often difficult to predict due to its complexity and high dynamics. This research aims to develop a model for predicting the success of software projects, particularly B2C e-business software in Indonesia, utilizing a machine learning approach. This study involved 28 variables that affect the success of software projects obtained from previous research. The dataset was compiled from the historical records of software projects from various software development companies in Indonesia. The predictive model was developed using Support Vector Machine and Artificial Neural Network algorithms, with hyperparameter tuning performed via Grid Search. The modelling process includes the pre-processing stage of data, which involves synthetic data generation due to inadequate data collection, as well as the application of several dataset mining techniques (SMOTE, ADASYN, SMOTE Tomek Links, and ADASYN Tomek Links). Additionally, model training and performance evaluation are conducted using a confusion matrix. The search for important features using the Shapley Additive Explanations method is also conducted to develop an automated recommendation system based on key factors that require improvement. The results showed that the SVM model with Grid Search tuning of hyperparameters in the SMOTE Tomek Links data test yielded the best performance, with an accuracy of 87.8%, demonstrating the significant potential of machine learning in identifying project success factors from the early stages. This study contributes to the development of decision-support tools for B2C project managers in Indonesia by providing accurate early predictions and interpretable recommendations.