RESEARCH01-HL_DMT_DATA-RESULTS

The included data, results, and models support a research project about a decision-making tool for the determination of the distribution center (DC) location in a humanitarian logistics network. The R programming language was utilized for both the generation of the data and the calculation of the results.

It includes lists of 70,000 synthetic random undirected networks with 15 nodes per network with a density between 0.18 and 0.91 and weights between 5 and 520 kilometers. This data was used to calculate various Centrality Indices (CIs) such as Degree, Strength, Closeness, Harmonic, Betweenness, Eigenvector, Laplacian and Subgraph, and the total distribution distance as a Traveling Salesperson Problem (TSP) with the support of the Farthest Insertion Heuristic Algorithm. The list of networks is split into three parts. In the list of networks for training (pa: 50,400), validation (pb: 12,600), and testing (pc: 7,000). These datasets are available in folder (01).

From each of the three parts of the network list, we compute the CIs and TSP solutions using each node from each network as a starting point (a potential DC facility). We then label each node of each network according to the first and second lowest total distribution distance as SELECT (1, sl, or SL) or REJECT (0, ns, or RJ) a node as a DC. Then we combine the CIs and TSP solutions dataset with the decision labels, and we result in an imbalanced dataset with a minority class SELECT. These datasets are available in folder (02).

Now, with proper processing of the datasets, we can model different approaches to a Binary Classification Problem to identify the appropriate node as a DC. In folder (00), there is the best Deep Neural Network model we use in our decision-making tool after conducting analysis and numerous experiments, with part of these results presented in folders (03 and 04). In addition, in the folders (00, 05, and 06), there are Decision Tree models and results that we used to compare their quality against the DNN model.