REVISTA DYNA NEW TECHNOLOGIES

Key words:

Redes neuronales, proyectos TIC, proceso de ventas, predicción de éxito, resultados de negocio, Neural networks, ITC projects, sales process, prediction of success, business results

Article type:

ARTICULO DE INVESTIGACION / RESEARCH ARTICLE

Section:

RESEARCH ARTICLES

ABSTRACT:
This research shows that the use of neural networks can help to predict the success of projects in a large ITC company, improving their business outcomes and its sales process. The research was carried out in a company with an annual average of 72,350 ended projects, which were managed using different methodologies. Data related to a period of 9 years were used. A successful project, according to the company's guidelines were those that had 0 unplanned impacts, a maximum deviation of 4% over the estimated budget, a maximum deviation of 2 weeks from the expected end date and a evaluation by the customer in the range [7, 10]. These variables were calculated separately using 4 RNA.
As a neural network model, a High-Order Multi-Layer Perceptron (MLP) was used, wich had 1 input layer with 16 nodes, 3 hidden layers with 11 nodes and 1 output layer with a node, as well as with Sigmoid, Rectified Linear Unit, Hyperbolic Tangent, and Arc-hyperbolic sine activation functions in the hidden layers. The activation function used in the input and output layers was the Sigmoid function. According to the established requirements, the RNA should achieve a Mean Square Error (MSE) of less than 0.00025 and a training time of less than 1.50 hours, with a learning rate of less than 0.1. Several sizes of the training set were selected in order to carry out 145 experiments with each activation function and training set. The best results were achieved using the Rectified Linear Unit activation function in the hidden layers, which achieved the goals for the four variables. The solution was deployed in an operational environment for one year, an increase in the percentage of projects that met the success goals was achieved. A survey was filled out by all the people related to the solution and involved in the sales process, obtaining an average score of 9.2. The questions were scored between 1 (worst value) and 10 (best value)

Keywords: Neural Networks, TIC projects, Sales Process, success prediction, business results