This article details the Tech Specs on your prediction report after uploading your dataset and training your ML model. This page gives technical information and detailed insights into the algorithm, model’s performance, accuracy and other advanced model metrics

First you get an overview of the general model metrics such as -

Algorithm Name: Best performing algorithm chosen by the platform

Accuracy (classification): Metric to evaluate your model’s performance for classification tasks

R2 score (regression): Metric to evaluate your model’s performance for regression tasks

Performance: indicates how the model is performing, depends on the accuracy/R2 score values and can go from Poor to Great

Next you can have a detailed view of your uploaded dataset again here. The extra information here is that you can also check for the number of rows and columns dropped, number of outliers dropped as these happen during the preprocessing of the data in the backend. Moreover you can also check columns used as the identifier column, Prediction column, feature columns used for training your model and the feature columns that were dropped (either you dropped manually or dropped by the platform because they did not meet the necessary requirements)

Next we have the Advanced Model Metrics section. This section gives the details of the algorithm itself after it is trained and the test set is run through it . For example, during a regression analysis, we have the metrics such as Training RMSE, Validation RMSE, Testing RMSE, R2 score, Mean Squared Error, Mean Absolute Percentage Error, Loss Function used by the algorithm, etc..

The corresponding values are shown on the Value column and the last column reflects on whether a value should be lower or higher to be considered as a good value. For example we need the RMSE to be lower and the R2 score to be higher. You have the option to download the advanced model metrics using the download icon on the top right.

For a classification task, we have the metrics such as Training, Validation and Testing accuracies, Precision, Recall, Area Under Curve, Loss Function and Learning rate of the algorithm, etc..

The last section on Tech Specs is the other machine learning models that were tested. Typically, the way our platform works is that once the dataset is ready to be fed into a model, in the backend, it runs multiple models with different hyperparameter combinations

We have 6 main algorithms for both classification and regression and in the backend we run 10000+ algorithms (taking into account all the hyperparameter combinations of the 6 different algorithms) while generating a prediction report. Finally we show the results of the top 2 best performing hyperparameter combinations of each algorithm, resulting in 12 best accuracy values in case of classification tasks and 12 best R2 scores in case of regression tasks . We show the results of each such model and their corresponding performance metric under “Value”

For classification we have the following algorithms:

Logistic Regression

Perceptron

Random Forest Classifier

XGBoost Classifier

Voting Classifier

LightGBM Classifier

Similarly, for regression we have the following algorithms:

Random Forest Regressor

XGBoost Regressor

Voting Regressor

Stacking Regressor

Gradient Boosting Regressor

LightGBM Regressor

After running all these algorithms in parallel, the best performing algorithm is chosen by the platform. The best performing algorithm normally is the one which has the best accuracy/R2 score among all the models tested. But our platform also checks for all the other advanced metric values to decide on the best algorithm. Thus the chosen algorithm may not always have the best possible accuracy/R2 score but is the best performing algorithm overall when compared with others

You also have the option to rerun the data with a model of your choice after getting the first prediction report. You can just click on the “Re-run with this model” for the corresponding algorithm that you are interested in running your data and get a prediction report that overrides the previous one