The "INDIAai" (National AI Portal of India) portal provides weekly comprehensive articles highlighting the research contributions made by universities and colleges in India.

We aim to offer thorough reporting on the AI research contributions made by a single institution every week. This series allows researchers and students to provide concise explanations of their work.

With more than 20,000 students and 1,500 faculty members, SRM Institute of Science and Technology (formerly SRM University) is among the best universities in India. It provides various undergraduate, graduate, and doctorate programs in engineering, management, medicine and health sciences, science, and humanities. 

Likewise, for the Semester Abroad Program, Thirty-five international universities, including MIT, Carnegie Mellon, UC Davis, Warwick, and Western Australia, funded more than 150 students.

Let us explore the top AI research contributions from SRM Institute of Science and Technology, College of Engineering and Technology, Ramapuram campus, Chennai.

Deep recurrent neural network-based Aquila optimization-based online shaming emotion analysis

The researchers B Aarthi Balika and J Chelliah state that "People now have a wealth of options for freely expressing their opinions, thanks to the rise of social media and online platforms. Consequently, this offers many opportunities for developing intelligent systems; yet, certain individuals exploit these platforms by defaming, intimidating, or mistreating others."

They say, "Emotional analysis of online shaming is required to prevent the attackers' unfavourable outcomes. Much research is carried out in this area to categorize shameful statements and those that are not to pinpoint the shamers. Conversely, the ideal classification is still up for debate. Consequently, we provide a new deep learning methodology called the deep recurrent neural network based classifier that classifies the comments." 

The researchers said, "We employed the Aquila optimization method, which enhances shaming categorization based on category, to increase classification accuracy. The performance indicators are identified and compared to previous studies for experimental analysis. According to the performance analysis, the suggested method works better than all other approaches."

Hybrid Whale Archimedes Optimization–based MLPNN model for soil nutrient classifcation and pH prediction

The researchers Prabavathi Raman and Balika Joseph Chelliah state that "In the agricultural industry, farmland quality and production are enhanced by environmental conditions and soil fertility. A novel classification and prediction model for soil nutrient potential and pH levels is proposed." 

They say that "Soils with the following characteristics are collected from the villages: phosphorus (P), organic carbon (OC), boron (B), and potassium (K). The objectives of the village-wise soil fertility prediction and classification model are to increase productivity, lessen the use of toxic fertilizers, improve soil health, and improve environmental quality. To improve the classification performance on the validation data, the suggested model integrates the Hybrid Whale Archimedes Optimization (HWAO) algorithm with the Multilayer Perceptron Neural Network (MLPNN) model." 

They added, "The soil nutrient prediction and classification model is validated using the Marathwada dataset, and a range of evaluation metrics are employed, including cross-validation accuracy, Area Under Curve (AUC), accuracy, Mean Squared Error (MSE), G-mean, precision, specificity, and sensitivity." 

The researchers further said, "This paper's comparative analysis demonstrates that the suggested HWAOMLPNN achieved a higher classification accuracy of 98.1%, cross-validation accuracy of 98.3%, and soil nutrient classification accuracy of 97.9%. With the help of the suggested model, soil nutrients and pH levels can be precisely classified. It can significantly improve soil health, lower the need for toxic fertilizers, improve the quality of the surrounding environment, and eventually increase profitability in the agricultural industry."

Analysis of demand forecasting of agriculture using machine learning algorithm

The researchers A Senthilselvi, T P Latchoumi, and Balika J Chelliah state, "The state of India was built on river deltas with suitable agricultural land and rich soil. As of 2019, the country's total surface area comprises more than 40% agricultural fields, mainly cropland. Furthermore, the country's Gross Provincial Product (GPP) is less than 3% of the earnings from the agricultural sector. Although manufacturing has emerged as the nation's primary economic activity, accounting countries account for 50% of GPP's total income." 

They added, "The study's goal is to determine how to increase farming supply chain networks' financial profitability and efficiency in the following ways: 

(1) Setting national goals for data on zonal groupings following projections and assessments that impact agricultural production and distribution. 

(2) Producers' risk can be minimised by providing consumers with various factory and industrialization options based on market analysis. 

(3) By standardizing the relationship between bankers and producers and centralizing land information through the structure, insurance costs can decrease, and bank borrowing can be reduced. 

(4) To stabilize the agricultural sector, consider regulating the Public Distribution System (PDS) for the security stock and exploring neighbouring possible production destinations." 

The researchers added, "This research presents a unique machine learning target prediction method to strengthen farmer-banker ties by informing farmers about market target products and centralizing information about recent government initiatives. The ML system for crop prediction was developed to increase agricultural revenue."

Mango leaf disease identification and classification using a CNN architecture optimized by crossover-based levy flight distribution algorithm

The researchers M Prabu and Balika J Chelliah state, "Researchers have lately employed a variety of computer-aided and machine learning techniques to classify illnesses affecting mango leaves. Nevertheless, it has been noted that these methods perform somewhat below par and that these issues can be linked to issues with overfitting, greater feature dimensionality, computational complexity, and poor feature quality." 

They say, "We suggested a unique paradigm for the disease classification of mango leaves to address these problems. The photos were shot in Andhra Pradesh, which has the most mango farms in India. The four stages of the suggested framework are the stage of data preparation, the stage of feature selection, the stage of learning and classification, and the stage of performance evaluation. From the healthy and unhealthy categories, we chose 380 photos."

They further said, " Various data augmentation methods enhance generalization and avoid overfitting. A convolutional neural network with crossover-based levy flight distribution is then used for improved feature selection. Additionally, the pre-trained MobileNetV2 model is employed during the learning phase, and at the end of the model, support vector machines are utilized to classify diseases. The experimental findings show that this strategy performs better in classification than other state-of-the-art approaches."