Workflow OverviewAccess Link is a powerful tool designed to facilitate the training and testing of Artificial Neural Networks (ANN) through an intuitive user interface. This application streamlines the process of building and evaluating ANN models by providing a range of user-friendly features. By leveraging the capabilities of, users can effortlessly upload datasets, configure ANN architecture, select hyper-parameters, perform cross-validation, and assess model performance. Let's explore the functionalities of this tool in detail. User Interface (UI) Panel:The UI panel of consists of several interactive components that enable users to input their preferences and interact with the tool efficiently. The following steps outline the key features of the UI: 1. Dataset Upload:Users are prompted to upload their dataset in CSV format. This ensures seamless integration of user data into the application. 2. Dependent Variable Selection:Using the select input feature, users can conveniently specify the dependent variable of their dataset. This allows for accurate modeling of the target variable. 3. Independent Variable(s) Selection:With the ability to select multiple independent variables, users can easily define the input features for their ANN model. This flexibility enhances the versatility of the tool. 4. ANN Architecture Specification:Users are prompted to determine the size (number of neurons) for the ANN architecture. This parameter influences the complexity and performance of the model. 5. Hyperparameter offers the option to set two essential hyper-parameters including maximum number of weights, maximum number of iterations, and Grid-Search hyper-parameters’ tuning Number of Neurons and Weight Decay based on RMSE error. This functionality enables users to explore various parameter combinations for optimizing the model performance to fine-tune the ANN models. 6. also offers the flexibility to select optimization algorithms for the training of the MLP-ANN model. Users can choose from a range of available algorithms to customize their modeling process. The following algorithms can be selected by users:
  • Port: The default algorithm, which utilizes the BFGS algorithm and a linear search.
  • BFGS: The Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm.
  • LBFGS: The Limited-memory BFGS algorithm.
  • CG: The Conjugate Gradient algorithm.
  • CGS: The Conjugate Gradient Squared algorithm.

By default, employs the "port" algorithm, which combines the BFGS algorithm and a linear search. 7. Cross-Validation Folds:To ensure robust model evaluation, users can specify the number of folds for cross-validation using a slider input. This allows for comprehensive assessment of the model's generalization capabilities. 8. Random-Sampling Train/Test Sets:Through a slider input, users can determine the proportion of data allocated for the training and testing sets. This facilitates the evaluation of model performance on unseen data. 9. Train Button:A "Train" button is provided to initiate the training process for the ANN model. Clicking this button executes the training algorithm and optimizes the model using the specified parameters. 10. Test Button:A separate "Test" button enables users to evaluate the performance of the trained ANN model on the testing set. This step assesses the model's ability to generalize to unseen data. 11. Downloadable Performance allows users to download performance metrics for both the training and testing sets. These tables provide insights into the model's accuracy, including:
  • Coefficient of Determination (R-squared), 
  • Mean Squared Error (MSE)
  • Root Mean Squared Error (RMSE)
  • Mean Absolute Error (MAE)
  • Mean Absolute Percentage Error (MAPE)
 12. Downloadable Performance Bar Plots:In addition to the performance tables, provides a unique feature that enhances the visualization of performance metrics. Users can download downloadable performance bar plots that integrate the results of performance metrics into a single plot. This consolidated visualization offers a comprehensive overview of the model's performance across multiple metrics, allowing users to quickly identify strengths and areas for improvement. By downloading these plots, users can easily share and analyze the performance of their ANN models. 13. Tabular Data Visualization:The uploaded dataset is rendered as a visually appealing datatable in the tab-panel named "Datatable." This feature enhances data exploration and understanding.
14. Model-Tuning includes an interactive visualization feature under "ANN Tuning" tab that illustrates the process of tuning hyperparameters during the grid-search optimization. The visualization provides insights into the relationship between the Weight Decay, RMSE, and the Number of Neurons in the ANN model. By visualizing this tuning process, users can gain a deeper understanding of the impact of different hyper-parameter values on the model's performance. 15. Performance calculates and displays the error results separately for the train and test sets. The tab-panel named "Performance" presents performance metrics such as R-squared, MSE, RMSE, MAE, and MAPE, enabling users to gauge the effectiveness of their ANN models. 16. Un-Boxing Blackbox of ANN (Variable Importance Visualization) generates an interactive bar-plot visualization of variable importance based on Olden and Garson algorithms. This allows users to identify the significant features contributing to the model's predictive power. 17. Neural Network Architecture performs an interactive visualization of the neural network architecture plot that illustrates the connections and layers within the ANN model. This visualization provides users with a comprehensive understanding of the network's structure and aids in the interpretation of the model's behavior.