Table 3. Accounting problem types and...

Table 3.

Accounting problem types and available ML techniques

Description of the learning problem	Solutions to learning problem	ML techniques	Source
Process 1: Translation of manual and electronic documents into accounting information
Task 1.3 Document features extraction
Feature extraction is an important process of obtaining relevant data before the classification of images. This process can be improved using ML to perform deep feature extraction	Classification	Deep convolutional neural network	Goussies, Ubalde, Fernandez, et al. (2014), Tarawneh et al. (2019)
Task 1.4 Document type recognition and classification
Image classification can detect the document type, which can be enhanced using ML	Classification	Convolutional neural networks New document class: k- nearest neighbour Similar known documents: support vector machine	ABBYY Technologies (2017), Oquab, Bottou, Laptev, et al. (2014), Sorio (2013), Sorio, Bartoli, Davanzo, et al. (2010), Witten, Frank, Hall, et al. (2016), Khan (2019), Tarawneh et al. (2019)
Irregular document layout classification using NLP combined with ML to train the system to process flexible or irregular document layouts	Classification	Convolutional neural networks	Chen et al. (2015)
Text classification is used to classify text using statistical and semantic text analysis	Classification and Clustering	Classification: Naïve Bayes Clustering: Parallelisation MapReduce k-nearest neighbour Semi-supervised clustering	Zhang et al. (2015), ABBYY Technologies (2017), Du (2017), Desai et al. (2021)
Task 1.6 Validation of document data:
Validation of document information can use ML to determine whether the extracted data from the document is correctly classified	Classification	Naïve Bayes Support vector machine	Larsson and Segerås (2016)
Removing duplicate entries and linking documents may be achieved using approximate string matching and ML for string classification	Classification	Naïve Bayes Decision trees Support vector machine Artificial neural network	Amtrup, Thompson, Kilby, et al. (2015), Larsson and Segerås (2016), De Leone and Minnetti (2015), Samoil (2015), Winkler (2014)
Process 2: Reconciliation of financial information
Task 2.3 Matching
Matching records or record-linkage have been performed using various ML techniques	Classification	Naïve Bayes Decision trees Support vector machine Artificial neural network	Chew and Robinson (2012), Samoil (2015)
Process 3: Preparation of management accounts
Task 3.3 Account allocation
Account allocation may be performed by incorporating ML, which learns to predict the account allocation based on probability and can recommend which accounts to post to	Classification and clustering	Classification: Naïve Bayes Clustering: K-means clustering Random forests	Bengtsson and Jansson (2015), Brady et al. (2017), SMACC (2017), Takaki and Ericson (2018)
Task 3.7 Report generation
Error detection in financial data and fraud detection can be performed by incorporating ML to identify irregularities in data sets	Classification; outlier detection and clustering	Classification: Bayesian belief network and a decision table Naïve Bayes hybrid model Outlier detection: Association rules	Ahmed et al. (2016), Alpar and Winkelsträter (2014), Hajek and Henriques (2017), Kokina and Davenport (2017)
		Clustering: K-means clustering Self-organising maps
Task 3.8 Report descriptions
Report descriptions may incorporate ML techniques in natural language generation technologies to enable a reasoning process to be applied to the reported data to produce required explanations in natural language	Prediction	Conditional random fields	Gardent and Perez-Beltrachini (2017), Lafferty et al. (2001), Yseop (2017)

Description of the learning problem	Solutions to learning problem	ML techniques	Source
Process 1: Translation of manual and electronic documents into accounting information
Task 1.3 Document features extraction
Feature extraction is an important process of obtaining relevant data before the classification of images. This process can be improved using ML to perform deep feature extraction	Classification	Deep convolutional neural network	Goussies, Ubalde, Fernandez, et al. (2014), Tarawneh et al. (2019)
Task 1.4 Document type recognition and classification
Image classification can detect the document type, which can be enhanced using ML	Classification	Convolutional neural networks New document class: k- nearest neighbour Similar known documents: support vector machine	ABBYY Technologies (2017), Oquab, Bottou, Laptev, et al. (2014), Sorio (2013), Sorio, Bartoli, Davanzo, et al. (2010), Witten, Frank, Hall, et al. (2016), Khan (2019), Tarawneh et al. (2019)
Irregular document layout classification using NLP combined with ML to train the system to process flexible or irregular document layouts	Classification	Convolutional neural networks	Chen et al. (2015)
Text classification is used to classify text using statistical and semantic text analysis	Classification and Clustering	Classification: Naïve Bayes Clustering: Parallelisation MapReduce k-nearest neighbour Semi-supervised clustering	Zhang et al. (2015), ABBYY Technologies (2017), Du (2017), Desai et al. (2021)
Task 1.6 Validation of document data:
Validation of document information can use ML to determine whether the extracted data from the document is correctly classified	Classification	Naïve Bayes Support vector machine	Larsson and Segerås (2016)
Removing duplicate entries and linking documents may be achieved using approximate string matching and ML for string classification	Classification	Naïve Bayes Decision trees Support vector machine Artificial neural network	Amtrup, Thompson, Kilby, et al. (2015), Larsson and Segerås (2016), De Leone and Minnetti (2015), Samoil (2015), Winkler (2014)
Process 2: Reconciliation of financial information
Task 2.3 Matching
Matching records or record-linkage have been performed using various ML techniques	Classification	Naïve Bayes Decision trees Support vector machine Artificial neural network	Chew and Robinson (2012), Samoil (2015)
Process 3: Preparation of management accounts
Task 3.3 Account allocation
Account allocation may be performed by incorporating ML, which learns to predict the account allocation based on probability and can recommend which accounts to post to	Classification and clustering	Classification:Naïve BayesClustering:K-means clusteringRandom forests	Bengtsson and Jansson (2015), Brady et al. (2017), SMACC (2017), Takaki and Ericson (2018)
Task 3.7 Report generation
Error detection in financial data and fraud detection can be performed by incorporating ML to identify irregularities in data sets	Classification; outlier detection and clustering	Classification: Bayesian belief network and a decision table Naïve Bayes hybrid model Outlier detection: Association rules	Ahmed et al. (2016), Alpar and Winkelsträter (2014), Hajek and Henriques (2017), Kokina and Davenport (2017)
		Clustering: K-means clustering Self-organising maps
Task 3.8 Report descriptions
Report descriptions may incorporate ML techniques in natural language generation technologies to enable a reasoning process to be applied to the reported data to produce required explanations in natural language	Prediction	Conditional random fields	Gardent and Perez-Beltrachini (2017), Lafferty et al. (2001), Yseop (2017)

Source: Compiled by author from multiple sources as indicated above

[ViewLarge]