OCL Based Approach for Sustainable ML Model Development

. It became a bottleneck for the Machine Learning (ML) researchers to select/develop a sustainable model for a particular problem. Hence, there is a need for an approach to prepare a model with all constraints of the software system. The proposed approach is based on Object Constraint Language (OCL) which is a declarative language for writing constraints on software artifacts, it is widely used for effective representation of Functional Requirements (FR’s) and Non-Functional Requirements (NFR’s). In the proposed system, the paddy leaf disease identification system is considered and proposed Leaf Identification Constraints (LIC) and Leaf Disease Identification Constraints (LDIC) based on OCL, for the proposed constraints the Convolutional Neural Network (CNN) is chosen, as it can handle diverse range of input data and large volume of concurrent requests. To satisfy other constraints of the system, the Auto encoders are used along with CNN and the input data was take in the form of thermal imaging. This system was evaluated with test data and validation data and obtained the accuracy of 90.6%. And 84.8 was attained by earlier researchers before this approach.


Introduction
The Machine Learning uses mathematical models and statistical analysis to enables computers to learn and improve their performance without explicit program.In the modern era the machine learning techniques became an emerging solutions for the real-time problems.As these ML models are rowing day by day, it become a tough task for the model developers to identify a suitable model or modifying the existing model or to developing a new model.*Corresponding author: ramesh680@gmail.comHence, this research proposed a mythology for solving the above mentioned problem.It is based on OCL [1] to identify the constraints of problem and identifying the suitable model or finding a model and modify according to the requirements of the system.Basically the OCL is a constraint based language provides a way to express precise and unambiguous constraints on Software artifacts.These constraints can be applied to ML systems during its model preparation.
The proposed approach provides twenty constraints for leaf identification and twenty constraints for leaf disease identification.Based on these OCL constraints the ML model is prepared, and tested.

Literature Survey
In the field of machine learning, most of the developers are exercising different models and fining the suitable one at the end for the given problem.Chiu, focused on different regression models and compared the performance of the differential evolution algorithm to that of the genetic algorithm [2].Using three benchmark datasets, the results show that the differential evolution method beats the genetic algorithm.The algorithms' performance is evaluated using R-squared, MAE, and RMSE.But this approach may not suitable for other algorithms like clustering and classification.Differential evolution (DE) algorithms [3] provides the most general method for optimizing the large and challenging problems.It was experimented with agriculture problems.PK Ram [4] have suggested the work based on the evolutionary algorithm with the multilayer perceptron for gene analysis using microarray-based health care data.Analyzed genes are responsible for the disease prediction.Here, vectors are designed in efficient way and the fitness function is derived to measure the quality of each vector.They also have designed the genetic algorithm based model to select the good features for disease analysis.Here, chromosomes are designed and to evaluate the fitness of chromosomes, new fitness function is derived.Basically, the fitness function is evaluated using the conflicting objective function.Afterwards, machine learning classifiers are used to measure the accuracy of selected feature subsets [5].
Taskin [6] proposed feature-selection technique for hyperspectral image analysis for handling the problems of classification and dimensionality reduction.This research improved the classification accuracy, stable feature selection, and effective computational performance are all attributes of the approach.Kumar [7] proposed a system that leverages CNN techniques to analyze medical images and accurately detect the presence of brain tumors.The study demonstrates the effectiveness of deep learning in medical image analysis and highlights the potential of cloud-based solutions for efficient and secure tumor detection.Sasank [8] proposed a novel framework that combines CNN and LSTM machine learning approaches to improve evaluation and enable systematisation in diverse cervical spondylosis-related applications.In [9,10] demonstrated how deep learning techniques, in especially CNNs and RNNs, have the ability to solve privacy and security problems.It analyses the use of DL approaches in attack classification, highlights the significance of intrusion and malicious detection at the node or peer level, and promotes the use of modern IDS systems to reduce harm and false alarms.In [11] the authors explored a novel approach in the form of an enhanced communication paradigm, introducing the Energy Aware Smart Home (EASH) framework.Within this study, we delve into the investigation of communication failures and various types of network attacks occurring in the context of EASH.By harnessing the power of machine learning techniques, we effectively distinguish the sources of abnormalities within the communication paradigm.
JS Kumar [12] proposed a method for analysing patient data using eXtreme Gradient Boosting (XGB) machine learning algorithm to provide diabetes risk predictions.This research highlights the usefulness of AI in healthcare by displaying machine learning's potential in diabetes screening and prevention.Thulasi [13] Providing the projections for Bitcoin prices based on previous data, allowing participants to make wise choices and possibly generate extra income.It is stressed how crucial market analysis and capitalization are, and it is acknowledged that Bitcoin is a substantial digital store of value.Ram et al [14], have proposed the novel genetic algorithm based on the autoencoder with ensemble classifiers for imbalanced health care data.Initially, the imbalanced data is balanced using the novel approach called as GAAE method.Here, genetic algorithm is evaluated through the autoencoder.Each chromosome of GA represents as an autoencoder.To measure the quality of chromosome, an error function is also designed by the authors.After balancing the dataset, feature selection is performed using the correlation coefficient approach.They proposed the feature clustering strategy using particle swarm optimization (PSO) technique for disease analysis.Here, the particles are designed in an efficient manner.Afterwards, the clustering scenario is developed using the correlation coefficient approach during the optimization of PSO process [15].Accurate classification and identification of plant diseases are achieved through the implementation of computer-based image recognition schemes.An advanced classification approach based on Back Propagated Artificial Neural Networks (ANN) is employed to implement feature-based matching operations in artificial intelligence [16][17][18][19][20].
OCL is the one of the declarative language for writing constraints on the FR's and NFR's of a software system at design level.Bolognesi [21] emphasizes the advantage of using constraint-oriented approach for system decomposition.This approach is consistent with object oriented reasoning added advantages in terms of enabling conditions and validation time.

OCL Based Constraints
The declarative language OCL is recognized by Object Management Group (OMG) and used along with Unified Modeling Language for specifying the constraints for software systems.This research extending the use of OCL for ML model preparation.As a part of model preparation the paddy leaf disease identification problem is taken into the consideration.Here, there are twenty leaf identification constraints and twenty leaf disease constraints are considered and tabulated in Table1.The leaf disease description is tabulated in Table 2.

Images with invalid data type
Bacterial leaf blight 14.
Images with very low contrast Curvularia leaf spot 15.
Images with very high contrast Leaf scorch 16.
Images with low resolution Brown stripe 18.
Images with high resolution Rice leaf folder 19.
Images with low quality Rice leaf mite 20.
Images with high quality Magnaporthe oryzae Brown stripe A bacterial disease that can cause yellowing of the leaves, followed by the development of brown stripes that run along the length of the leaf blade.The disease can cause severe yield loss, particularly in wet conditions.18.
Rice leaf folder A pest that can cause the folding and rolling of the leaves, which can result in reduced photosynthesis and yield loss.The pest can also transmit viruses and other pathogens.19.

Rice leaf mite
A pest that can cause the yellowing and drying of the leaves, which can result in yield loss.The pest can also transmit viruses and other pathogens.20.

Magnaporth e oryzae
A fungal disease that causes blast symptoms on the leaves.It can also affect other parts of the plant, including the stem and grain, and can cause significant yield loss.
The OCL constraints are written for the paddy leaf diseases are tabulated in table 3. Here, for the give twenty diseases OCL expressions and Description is given.Table 4. represents the leaf identification constraints which includes the expression of the artefact, precondition, post-condition and its description of all the leaf identification constraints.Leaf must have at least one brown spot with a light-brown border, irregular shape, and width greater than 2 units, and at least one gray circular spot with length greater than 2 units.

17.
Brown stripe self.disease= 'Brown stripe' implies self.spots->exists(spot| spot.color= 'brown' and spot.border= 'dark' and spot.shape= 'striped') and self.spots->exists(spot| spot.color= 'gray' and spot.shape= 'circular' and spot.width> 2) Leaf must have at least one brown striped spot with a dark border and at least one gray circular spot with width greater than 2 units.18.

Results and Discussion
The OCL constraints for paddy leaf detection and paddy leaf disease detection are written from the requirements of the system.And from the constraint tables (i.e.Table 3    The Fig. 1 shows the one sample result identifying Hispa disease and Fig. 2 shows as identifying Leaf-Spot Disease.The Fig. 3 show, the proposed model identifying healthy leaf, and Fig. 4 shows the comparison of the accuracy with existing model.The proposed model produced satisfactory results when compared with existing model.

Conclusion and Future Directions
This research introduced OCL based ML model preparation approach in cost effective way.Based on OCL there were 40 constraints are generated for choosing a problem and prepared a model.This model gave satisfactory results the early approaches.This approach leads to accurate classification of training data, testing data and validation data.It maximizes the test coverage, accurate prediction.
In the feature the OCL can be promoted to another level of models like Deep Learning (DL) system.And specific applications like vehicular Internet of Thing (IoT), Health care IoT, Agriculture IoT, where there is a great need of ML models.
/doi.org/10.1051/e3sconf/20234300107272 430 4) the required model has been prepared and executed.Based on the trained data set the model is identifying the disease leaf and healthy leaf.

Figure 2 .
Figure 2. Image predicted as Leaf-Spot Disease

Figure 3 .
Figure 3. Image predicted as Healthy

Figure 4 .
Figure 4. Accuracy between existing and proposed models.

Table 1 .
Constraints for Leaf Identification and Leaf Diseases.

Table 2 .
List of twenty possible paddy leaf Diseases that can cause oval or diamond-shaped lesions with gray centers and brownish borders on the leaves.It can also affect other parts of the plant, including the stem and grain.2. that can cause circular or oblong lesions with dark brown centers and yellow borders on the leaves.The lesions can merge and cause extensive damage to the leaf tissue.17.
Brown spot A bacterial disease that can cause small brown spots with yellow halos on the leaves.The spots can merge and form larger patches, leading to leaf yellowing and premature senescence.3.Sheath blightA fungal disease that can cause brown lesions with yellow halos on the leaf sheaths, which can spread to the leaves and cause shredding of the leaf tissue.4.Bacterial leaf streakA bacterial disease that can cause yellowish-brown streaks on the leaves, which can become necrotic and cause lesions that can spread throughout the leaf blade.lesions on the leaves.The lesions may appear as gray, brown, or black spots with a dark ring around them.In severe cases, the lesions can merge and cause death of the leaf tissue.12.Sheath rot A fungal disease that can cause water-soaked lesions on the leaf sheaths, which can lead to the rotting and shredding of the

Table 3 .
Leaf Disease Constraints

Table 4 .
Leaf . Rice leaf self.disease= 'Rice leaf mite' implies Leaf must have at