AllMetrics - Calculating Multiple Performance Metrics of a Prediction Model

Provides a function to calculate multiple performance metrics for actual and predicted values. In total eight metrics will be calculated for particular actual and predicted series. Helps to describe a Statistical model's performance in predicting a data. Also helps to compare various models' performance. The metrics are Root Mean Squared Error (RMSE), Relative Root Mean Squared Error (RRMSE), Mean absolute Error (MAE), Mean absolute percentage error (MAPE), Mean Absolute Scaled Error (MASE), Nash-Sutcliffe Efficiency (NSE), Willmott’s Index (WI), and Legates and McCabe Index (LME). Among them, first five are expected to be lesser whereas, the last three are greater the better. More details can be found from Garai and Paul (2023) <doi:10.1016/j.iswa.2023.200202> and Garai et al. (2024) <doi:10.1007/s11063-024-11552-w>.

Last updated 11 months ago

1.78 score 2 dependents 387 downloads

DescribeDF - Description of a Data Frame

Helps to describe a data frame in hand. Has been developed during PhD work of the maintainer. More information may be obtained from Garai and Paul (2023) <doi:10.1016/j.iswa.2023.200202>.

Last updated 2 years ago

1.78 score 2 dependents 268 downloads

CEEMDANML - CEEMDAN Decomposition Based Hybrid Machine Learning Models

Noise in the time-series data significantly affects the accuracy of the Machine Learning (ML) models (Artificial Neural Network and Support Vector Regression are considered here). Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) decomposes the time series data into sub-series and help to improve the model performance. The models can achieve higher prediction accuracy than the traditional ML models. Two models have been provided here for time series forecasting. More information may be obtained from Garai and Paul (2023) <doi:10.1016/j.iswa.2023.200202>.

Last updated 2 years ago

1.78 score 2 dependents 299 downloads

WaveletML - Wavelet Decomposition Based Hybrid Machine Learning Models

Wavelet decomposes a series into multiple sub series called detailed and smooth components which helps to capture volatility at multi resolution level by various models. Two hybrid Machine Learning (ML) models (Artificial Neural Network and Support Vector Regression have been used) have been developed in combination with stochastic models, feature selection, and optimization algorithms for prediction of the data. The algorithms have been developed following Paul and Garai (2021) <doi:10.1007/s00500-021-06087-4>.

Last updated 2 years ago

1.48 score 1 dependents 223 downloads

RankPCA - Rank of Variables Based on Principal Component Analysis for Mixed Data Types

Principal Component Analysis (PCA) is a statistical technique used to reduce the dimensionality of a dataset while preserving as much variability as possible. By transforming the original variables into a new set of uncorrelated variables called principal components, PCA helps in identifying patterns and simplifying the complexity of high-dimensional data. The 'RankPCA' package provides a streamlined workflow for performing PCA on datasets containing both categorical and continuous variables. It facilitates data preprocessing, encoding of categorical variables, and computes PCA to determine the optimal number of principal components based on a specified variance threshold. The package also computes composite indices for ranking observations, which can be useful for various analytical purposes. Garai, S., & Paul, R. K. (2023) <doi:10.1016/j.iswa.2023.200202>.

Last updated 9 months ago

1.00 score 476 downloads

tsLSTMx - Predict Time Series Using LSTM Model Including Exogenous Variable to Denote Zero Values

It is a versatile tool for predicting time series data using Long Short-Term Memory (LSTM) models. It is specifically designed to handle time series with an exogenous variable, allowing users to denote whether data was available for a particular period or not. The package encompasses various functionalities, including hyperparameter tuning, custom loss function support, model evaluation, and one-step-ahead forecasting. With an emphasis on ease of use and flexibility, it empowers users to explore, evaluate, and deploy LSTM models for accurate time series predictions and forecasting in diverse applications. More details can be found in Garai and Paul (2023) <doi:10.1016/j.iswa.2023.200202>.

Last updated 1 years ago

1.00 score 132 downloads

MissingHandle - Handles Missing Dates and Data and Converts into Weekly and Monthly from Daily

Many times, you will not find data for all dates. After first January, 2011 you may have next data on 20th January, 2011 and so on. Also available dates may have zero values. Try to gather all such kinds of data in different excel sheets of a single excel file. Every sheet will contain two columns (1st one is dates and second one is the data). After loading all the sheets into different elements of a list, using this you can fill the gaps for all the sheets and mark all the corresponding values as zeros. Here I am talking about daily data. Finally, it will combine all the filled results into one data frame (first column is date and other columns will be corresponding values of your sheets) and give one combined data frame. Number of columns in the data frame will be number of sheets plus one. Then imputation will be done. Daily to monthly and weekly conversion is also possible. More details can be found in Garai and others (2023) <doi:10.13140/RG.2.2.11977.42087>.

Last updated 2 years ago

1.00 score 245 downloads

WaveletMLbestFL - The Best Wavelet Filter-Level for Prepared Wavelet-Based Models

Four filters have been chosen namely 'haar', 'c6', 'la8', and 'bl14' (Kindly refer to 'wavelets' in 'CRAN' repository for more supported filters). Levels of decomposition are 2, 3, 4, etc. up to maximum decomposition level which is ceiling value of logarithm of length of the series base 2. For each combination two models are run separately. Results are stored in 'input'. First five metrics are expected to be minimum and last three metrics are expected to be maximum for a model to be considered good. Firstly, every metric value (among first five) is searched in every columns and minimum values are denoted as 'MIN' and other values are denoted as 'NA'. Secondly, every metric (among last three) is searched in every columns and maximum values are denoted as 'MAX' and other values are denoted as 'NA'. 'output' contains the similar number of rows (which is 8) and columns (which is number filter-level combinations) as of 'input'. Values in 'output' are corresponding 'NA', 'MIN' or 'MAX'. Finally, the column containing minimum number of 'NA' values is denoted as the best ('FL'). In special case, if two columns having equal 'NA', it has been checked among these two columns which one is having least 'NA' in first five rows and has been inferred as the best. 'FL_metrics_values' are the corresponding metrics values. 'WARIGAANbest' is the data frame (dimension: 1*8) containing different metrics of the best filter-level combination. More details can be found in Garai and others (2023) <doi:10.13140/RG.2.2.11977.42087>.

Last updated 2 years ago

1.00 score 635 downloads

AriGaMyANNSVR - Hybrid ARIMA-GARCH and Two Specially Designed ML-Based Models

Describes a series first. After that does time series analysis using one hybrid model and two specially structured Machine Learning (ML) (Artificial Neural Network or ANN and Support Vector Regression or SVR) models. More information can be obtained from Paul and Garai (2022) <doi:10.1007/s41096-022-00128-3>.

Last updated 2 years ago

1.00 score 733 downloads

CompareMultipleModels - Finding the Best Model Using Eight Metrics Values

In statistical modeling, multiple models need to be compared based on certain criteria. The method described here uses eight metrics from 'AllMetrics' package. ‘input_df’ is the data frame (at least two columns for comparison) containing metrics values in different rows of a column (which denotes a particular model’s performance). First five metrics are expected to be minimum and last three metrics are expected to be maximum for a model to be considered good. Firstly, every metric value (among first five) is searched in every columns and minimum values are denoted as ‘MIN’ and other values are denoted as ‘NA’. Secondly, every metric (among last three) is searched in every columns and maximum values are denoted as ‘MAX’ and other values are denoted as ‘NA’. ‘output_df’ contains the similar number of rows (which is 8) and columns (which is number of models to be compared) as of ‘input_df’. Values in ‘output_df’ are corresponding ‘NA’, ‘MIN’ or ‘MAX’. Finally, the column containing minimum number of ‘NA’ values is denoted as the best column. ‘min_NA_col’ gives the name of the best column (model). ‘min_NA_values’ are the corresponding metrics values. ‘BestColumn_metrics’ is the data frame (dimension: 1*8) containing different metrics of the best column (model). ‘best_column_results’ is the final result (a list) containing all of these output elements. In special case, if two columns having equal 'NA', it will be checked among these two column which one is having least 'NA' in first five rows and will be inferred as the best. More details about 'AllMetrics' can be found in Garai (2023) <doi:10.13140/RG.2.2.18688.30723>.

Last updated 2 years ago

1.00 score 212 downloads