A fundamental change in the process of consumer decision-making in the digital world was predetermined by the active online interactions, the richness of customer experience, and the constant technological revolution. Every time people visit online shopping sites, social media, mobile apps, and online commercials, they leave a massive trail of behavioral information that can tell about latent likes and interests, plus buying desires [4]. This behavioral data has proven to be among the most useful resources in the modern marketing where organizations are keen to understand the consumers in order to manipulate their decisions appropriately. However, the complexity, flexibility, and rapidity of digital consumer data are far too complicated to be subjected to the normal methods of analysis. The unstructured features and the nonlinear patterns of consumer preferences that are occurring currently cannot be processed readily with the assistance of the statistical method and the addictive machine learning algorithms. The paper will answer these questions using deep learning models so as to make predictions with more accuracy and context when it comes to consumer decision-making.

In the digital marketing environment that is ever-changing, determining the decision of the consumer does not concern isolated purchasing behavior. Rather, it will need a complex view of browsing histories, search trends, sequences of clicks, reactions towards advertisements, feelings during product reviews, and demographic characteristics. The need to have the right predictive models has been growing even more compelling as companies strive to cut down on marketing costs, offer personalised promotions, enhance customer acquisition, and compete favourably even in the saturated markets. Consumer information is unprecedented with billions of digital users systems including Amazon, Instagram, TikTok, and online stores. However, most of this information cannot be put into practice without sophisticated algorithms, and marketers cannot gain insight at the level of single decision processes. Thus, deep learning application in marketing analytics can be viewed as a chance to convert lines of raw consumer traces into useful information [1].

Deep learning networks like the Convolutional Neural Networks (CNN), the Long Short-Term Memory networks (LSTM), and architectures are now very robust in modeling the multi-dimensional marketing data. In contrast to traditional models that incorporate a substantial amount of manual feature engineering, deep neural networks automatically learn hierarchical features using data, and thus they are capable of automatically identifying small yet influential behavioral patterns and timing behavior (that aid in affecting the decision to buy a product) [6]. An example is that LSTM networks are very useful in modeling long-term dependencies in clickstream sequences, and thu      s, the system can be conditioned to respond to repeated product views or extended browsing behavior with the probability of making a purchase. Instead, CNNs can be used to analyze written data contained in reviews, product descriptions, or social media interactions and extract sentiment and semantic signals, which influence the perception of users. The combination of these models into a hybrid system enables simultaneous consideration of temporal, contextual, and behavioral traits to be used by marketers which makes the decision-making more accurate.

Although the improvements have been made, the current research concentrates on limited datasets, including sentiment analysis or interfaces, which discloses a major gap in the multi-source, consumer-oriented prediction framework. Instead there have been numerous previous models that are based on structured concepts such as price sensitivity or simple demographic information that are not able to capture the intricacy of what modern digital behavior entails. The presence of the traditional marketing theory, as helpful as it might be, does not fit well into the data-driven world, where the decisions a user makes are based upon personalized recommendations, user-generated content, social approval and exposure algorithms. The present analytical paradigm should therefore accord equal civil role to deep learning among numerous multi-modalities of behavior to make it accurate and practical. This necessity is one of the significant points of the work motivation [3].

Finally, the research will also make an effort to layer academic theory and practice. Whereas much literature has been obtained to suggest that deep learning could make in someone the potential in isolated marketing jobs, there is decreasing literature that suggests the more detailed predictive framework that the capability to bring in an assemblage of varied behavioral antecedents to simulate the whole decision-making process. Through this work, there is a single architecture that responds to the reality in the marketing environments and the findings are useful to both the scholarly and the industrial applications. This research provides a platform to support marketing analytics at the next level, which is based on AI, personalization, and predictive intelligence by engaging the complexities of digital consumer behavior using an advanced deep learning framework [5].

Novelty and Contribution

This study presents a new multi-source behavioral data model based on a hybrid deep learning framework that will forecast consumer decision-making. This is in contrast to previous studies, which used individual datasets like clickstream logs or review sentiment analysis to generate a representation of consumer behavior but does not include the browsing sequences, demographic factors, social media activity, and textual sentiment indicators. The difference is in the fact that CNN based contextual feature extraction is combined with LSTM based temporal modeling so that the system will also learn sequential decision and underlying contextual patterns. Such two-fold design yields more precise and strong predictions as opposed to single deep learning models or conventional machine learning algorithms.

The main reason to do this work is the increasing in complexity of the digital consumer journeys decisions that are informed through a multi-platform exposure and immediate interaction. Current methods of analysis have been unable to portray these nonlinear and multi-layered behavior. To fill this gap, the study will use hierarchical feature learning in deep learning to read between the lines of behaviors, i.e. repeated product visits, time spent on product pages, social proof indicators, and the sentiment-driven response. This model therefore indicates the exact cases of marketing, which happen in reality because decisions have been formed by a combination of rational analysis, emotions, and browsing habit.

The fundamental aims of the research are three-folds:

• To create a predictive model based on deep learning that would be able to reproduce the complexity and non-linearity of consumer decision-making in the digital space.
• In order to incorporate the multi-source data on behavior that combines and affects purchase choices, the formation of an overall predictive data.
• The aims of the study include comparing CNN, LSTM, and the hybrid CNNLSTM models to determine which of them best predicts the decisions.

The significant contributions of the research are:

• A new hybrid deep learning system wherein CNN is integrated with LSTM in order to obtain spatial as well as time-related features of interaction between and among consumers.
• The multi-source behavioral data that covers click streaming sequences of behaviors, sentiment data, demographics and engagement measures which provides more predictive power than earlier single-source behavioral prediction systems.
• Empirical levels of performance showing that the hybrid model outperforms more through accuracy, recall, and AUC statistics.
• An efficient predictive system of individual marketing, client sectioning, and customized promotion enhancement.
• An explanation of operational weaknesses, such as the problem of interpretability of models and the data dependency, and the new directions, such as explainable AI, and privacy-sensitive learning.

In general, this research is novel and contributes to the advancement of science and practice as it will offer a complex, highly developed, and more practical deep learning system that raises the accuracy and intelligence of consumer decision-making prediction in digital marketing processes.

The study of the consumer decision-making process in online space has developed significantly over the last few years with the overwhelming tendency towards the algorithmic and data-based decision-making. Initial research in the area of study also depended on old statistical techniques like logistic regression, decision trees and clustering models in understanding issues that had effects on purchase intention. These methods gave good information on the demographic trends and simple predictors of behavior but suffered as they involved predetermined attributes, and linear interactions were not possible. These developments and the increasing volume and variety of digital consumer information meant that the traditional methods could no longer be used to model the complicated interaction between behavioural, emotional and contextual attributes in the formation of online choices. This breakthrough led to machine learning, and more recently, deep learning, as highly analytical systems capable of processing large amount of and unstructured consumer data.

In 2025 D. O. Hassan et.al., [7] introduced the technology improved in the field of computer science and consumer analytics, studies started employing machine learning algorithms including random forests, support vector machines and ensemble models to foresee consumer needs and categorism purchasing desires. These kinds of models showed much better performance compared to traditional statistical models by avoiding the need to hand-free feature engineer and support more complex structures. However, they were still limited in the context of application to a sampled data such as clickstream logs, search history and visitor behavior when time happens to be a major constraint. In addition, the majority of these methods were challenged in terms of manipulating unstructured data like text reviews, multimedia and social media interactions that are largely influential in the formation of consumer attitudes and behaviors.

Another critical mark of change in digital marketing analytics was the emergence of deep learning. Research on this industry began applying neural network models such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) network architectures to predict buyer behavior. The use of CNNs was quite handy when it comes to analysis of text based information with user reviews of products, sentiment analysis and description of items or goods being sold [8]. These models might bring to view semantic patterns that could not be determined using the classical algorithms, which enabled the need to comprehend the emotional values of the consumers and perceptions towards the products to be understood. In the meantime, LSTM models became popular due to their capacity to process consumer pathways over time, and thus dependencies in long sequences of clicks, searches, and product interactions. This ability has enabled the researchers to recreate the way consumers explore online space and the way their willingness changes with the course of time.

In 2025 F. E. Husseini et.al. [14] suggested the number of studies also examined neural networks that are hybrids, that is, a CNN, and LSTM layer added together in order to combine both contextual and behavioral elements. These mixed models proved to be the best in forecasting purchase intention, enjoys clicking through rates, and online conversions. Their advantage is that they can extract text or visual high level features with CNNs, and at the same time learn sequential dependencies with LSTM networks. The combination of this kind gives higher complete representations of the consumer behavior, especially in cases where the choice has been made under the influence of several data modalities. As a case in point, reading reviews, viewing short promotional videos, and repeatedly visiting a specific product page might influence consumer interest in a product. It is true that such interrelated behaviors can be modeled by the use of hybrid models with better forecasts than single methods models.

Along the same breath, sentiment analysis studies have had a huge influence on the consumer decision modeling. Even the emotion caused by the evaluation of products, remarks and references on social media have been found to be a strong analyzer of gratification and desire in the customer. Deep learning-based sentiment classification systems have managed to excel over the past systems of classification, whether rule-based or lexicon-based, by acquiring fine emotional signals in writing [9]. The addition of sentiment measures to prediction models has helped in achieving improved formulas in terms of differentiating the positive dispoiltonary behavior and real purchase intent. This has played a more critical role in the online setting where the opinions of customers play a very significant role in the purchase of something.

In 2025 K. Abrokwah-Larbi et.al., [2] proposed the other captivating course of study involves exploiting the phenomenon of attention engineering along with transformer-built models that have shown the ability to focus wretchedly on long-range dependence and contextual relations more effectively than other types of neural networks architecture did. By assigning some level of significance to various actions / interactions these models enable a more significant level of consumer behavior patterns to be entitled. Although they are not yet actively explored in terms of the digital marketing applications, they have massive opportunities to the future in the context of the predictive systems, which will be capable of dealing with much more complicated sequences of behavior.

Besides the effort of single models, current studies have also focused their interest on the use of multi-source data integration. The Internet consumers carry out communication using an extremely great variation of platforms, which consist of online stores, social network programmes, email marketing systems and cautious advertisements and creates a distributed foot julgment of behavior. The synergistic effect of the combination of these sources of data on predictive reliability has been found to be large with the provision of a holistic view of the consumer journey. The reason is that the click-stream records can tell the browsing patterns but with the addition of the sentiment data, demographic characteristics, engagement metrics, and time associated variables, may serve to form a further understanding of the consumer motivations. Studies have found that such multi-modal datasets have a great enhancement in the prediction of purchase intent, churn and personalized recommendation [15].

Despite these concerns, the existing literature shows that deep learning has a potential to be a game changer in predicting consumer decision making in online marketing. Even the trend in direction of richer data, multi-modal modeling and more complex neural structures are not the last efforts to fill in the menacing discrepancies between the realities of consumer behavior and useful marketing data. The present work builds on the premises of this and develops a hybrid deep learning model that takes into consideration sequential and contextual dimensions, which is a comprehensive and viable mode of functioning in the actual marketing environment.

The proposed algorithm of determining consumer choice given the deep learning model in the area of digital marketing is expounded as a multi-step analysis model to make user activities within the digital media complex to ease the user interaction. The plan opens with the creation of entire behavioral data in the form of click stream records, surfing data, demographic (age, gender, etc.), social media data, and customer sentiment information (reading user-generated text). Since the consumers can experience a multiplicity of online touchpoints during the decision-making process, a methodology of pooling these dissimilar kinds of data into a consistent structure permitting learning of profound characteristics is considered. The first phase aims at gathering crude information about ecommerce applications, advertising boards, web analytical software, and socially-published materials in the public domain. A specific behavioral information can be provided by each of the data sources, which makes sure that the end product of the work contains a complete range of cognitive, emotional, and contextual conditions that impact the purchasing behavior. The methodology does not consider these different signals in isolation but instead as related components of the decision process and so further modeling of time and contextual dependencies can be done [10].

After completion of data collection, methodology is switched to a period of massive preprocessing aimed at improving the quality of data, and the pre-processing of data as an input to the deep learning. Rawicted digital interaction data is usually filled with holes, inconsistent tab visitations, and garbage left by the activity of bots or accidental fast clicking. To handle these problems, a highly rigorous filtering process is implemented on the dataset by eliminating invalidated records and transforming several numerical characteristics using scaling methods. Features that are written like product reviews, search queries, and comment activity are cleaned in terms of tokenization, the elimination of stop-words, and representations using the forms of vectors embedding that transform natural language into a dense numerical representation that can be processed using neural networks. In case of behavioral sequences e.g., clickstream logs, data segmentation methods are used to transform continuous browsing logs into sequences of chronologically ordered events which characterizes a consumer as flowing through the sequence of exposure to potential purchase and final purchase. This is referred to as preprocessing, which is done to ensure that the dataset represents the true consumer behavioral pattern and is free of environmental distortions that can have adverse effect on model training.

The next phase of the methodology, after the preprocessing, is feature engineering and deep representation learning. In contrast to the traditional feature engineering, where feature selection is performed manually, with the assistance of deep learning architectures, a large portion of the relevant patterns can be auto-extracted with the help of this methodology. Nonetheless, some modifications are made in order to stimulate the learning process which can include but is not limited to encoding demographic variables into numerical values, consolidating engagement metrics into time-based activity summaries, and generating interaction intensity measures that can be used to demonstrate the extent to which a consumer is affiliated towards a product category or a marketing element. Multi modal data is constituted by combining these transformed characteristics with textual embedded and clickstream sequences. The reasoning process of this design is that consumer decision-making process can rarely be determined in one factor; rather, it happens as a result of emotional sentiment, repeated exposure, peer influence and relevance that is personalized. Consequently, a combination of extracted features with learned deep representations enables the system to learn latent drivers of decision making that might not be obvious in case of manual examination.

The approach successively leads to the phase of model development, where individual deep learning elements are also trained and then they are combined into a hybrid predictive model. The former is a Convolutional Neural Network that is mainly used with the textual and contextual data. This CNN network detects sentiment indicators, linguistic regularities, and semantic interactions in reviews of products, descriptions of ads, and posts made by other users, which tend to arise and potentially influence perception and intention. The second element is Long Short-Term Memory architecture which captures consumer browsing and engagement behavior sequentially. Since the decision-making process evolves over time, we utilize LSTM layers to learn long-run characteristics such as product visits, slow revisions of the site, slow transitions in the purpose of exploration of casual browsing. After the training of these two models separately with the respective modalities of the data, they are merged into a structure of a hybrid deep learning model. The architecture integrates both the nature of the LSTM with respect to time and the nature of CNNs with respect to space to enable a holistic perspective of consumer motivation to drive the content of the emotional context and consumer behavioral trajectories to consumer actions [11].

Training basis on the basis of model training with a batch processing procedure, early stopping procedures, and performance validation in numerous epochs utilizes the well-organized training process. The training data is separated into specific subsets to be used in the training process, validation, and testing to make sure that the model is able to generalize well to the unseen consumer behavior instead of sweating over the existing trends. The hybrid architecture is optimised sequentially to enhance accuracy and strength, such as hyperparametersoptimisation, optimization of network depth and dropout procedures that discourage overtraining. In an attempt to play up reliability, the performance evaluation measure(s) (accuracy, recall, precision, and AUC values) are calculated in every training cycle. The methodology is stable and repeatable as it includes the testing of every component of the model on the actual consumer interaction situation, including whether a user who has been interacting with a particular advertisement will eventually switch over to cart interaction or buy confirmation.

The last phase of the methodology is based on integration, prediction and deployment. After the validation of hybrid deep learning architecture, it will be implemented on new consumer data in real time to make a prediction on decision making. These forecasts could contain probability to buy, probability to drop-off, chances to participate in an advert or probability of turning over to the competitors. The outputs of the model are incorporated in marketing dashboards and decision-support systems used by the marketers to plan their strategy [13]. Dynamic personalization, focuses on specific promotions, and delivery of recommendations is supported by real-time prediction. The system changes the scores of predictions according to emerging shifts in customer behavior to remain relevant by constantly monitoring new parameters on the behavior of consumers in the market. The deployment stage also has optional features regarding model maintenance so that the system can self-train itself with newly gathering data so that it can enhance its prediction performance as time progresses. It ensures that the methodology is flexible and sustainable in the changing digital marketing conditions.

The Figure 1 shows the entire methodological workflow with the view of predicting consumer decision-making through deep learning models. First, there is data collection, in which various digital marketing data sets including clickstreams, demographics, and text data should be collected. This information is then preprocessed to remove noise, scale numerical values and change text into embedded numerical vectors after which feature engineering combines contextual, temporal and behavioral hints. The methodology then splits into two niche deep learning directions one CNN model that will attempt to derive sentiment and contextual cues out of text and an LSTM model that will attempt to derive sequential browsing and engagement patterns. These two models are then combined into hybrid CNNLSTM architecture that is capable of identifying the spatial and time decision-making patterns. The integrated model is then trained, then it is verified and run then deployed into actual marketing systems to bring about actual time consumer decision forecasts.

The outcomes of the constructed analytical structure prove a steady improvement in performance toward all the measured model sets with a specific focus on accuracy, precision, recall, and AUC numbers. Figure 2 shows that the CNN, LSTM, and Hybrid models have different learning trends that can be traced throughout the 10 training epochs with the Hybrid model on the top of the learning curve. The increase in accuracy is progressive, and it indicates that the fusion-based architecture can extract more discriminative data in the dataset.

FIG. 2: ACCURACY TREND OF CNN, LSTM, AND HYBRID MODELS

The presented performance trend is also consistent with the relative values depicted in Table 1, where z = Hybrid model demonstrates the highest accuracy of 0.88, which is higher in comparison with the two models (CNN and LSTM). This enhancement establishes the benefit of the hybrid representation of spatial and temporal characteristics in a unified representation, which results in more consistent pattern recognition in different test samples.

TABLE 1: PERFORMANCE COMPARISON OF MODELS

Upon additional study of precision and recall values, other facts about the reliability of the classifier can be valued. Figure 3 indicates that, as the conventional architectures lagged behind the Hybrid one, the precision of the former also indicated a corresponding good performance. These tendencies show that Hybrid model correctly classifies a larger percentage of relevant samples as well as provides wider range of true signals than CNN and LSTM do alone. These findings are confirmed with the corresponding metrics that are summarized in Table 1 such that both the accuracy and recall of the Hybrid framework are more than the other models. These experiments provide the correctness and validity of the proposed architecture in context of classification of inverted or imaginary data points using dirty or intersecting data points. The greater sensitivity that allows the system to reduce a false negative which is an inseparable aspect of critical tasks is also emphasized by the results of the better recall.

FIG. 3: PRECISION AND RECALL COMPARISON

Additional scrutinizing analysis of discriminative ability, was carried out through AUC which contrast the significance of the ROC analysis to added to a more general model comparison not simply on the basis on thresholds. The tests demonstrated that the value of ROC AUC differ significantly between Hybrid and CNN and LSTM with the highest AUC of 0.91 and 0.82 and 0.85 respectively; ad depicted in figure 4. This promotion signs a fantastic improvement in the categorization of positive and negative classes in every one of the probabilities levels.

FIG. 4: ROC AUC VALUES OF MODELS

Such values on Table 2 also indicate such performance results which also includes the F1 and the Log Loss comparisons. It is curious to observe that the Hybrid model has the smallest log loss (0.31), and therefore generates its outputs better calibrated with ground truth distributions. The same would be coupled with the highest F1 score of 0.84 that represents a superior balance among accuracy/precision and memory/recall- a property required in the implementation environment.

TABLE 2: EVALUATION METRICS ACROSS MODELS

The similar trends of the learning processes in all three diagrams provide evidence of the significance of multi-representation methods of learning in improving the credibility of a system. The fact that the Hybrid architecture allows integration of spatial convolution as well time repetition, makes the architecture become more dependable as far as input variations are concerned. CNN and LSTM, however, being efficient on their own, have been found to be ineffective, as far as they are supposed to identify cross-dimensional patterns simultaneously. This disparity is particularly well observed in Figure 1 with the LSTM closely tracking the CNN at the early epochs but slightly develops a deviation as the training goes on with a consequent lower eventual value of final accuracy. This divergence would suggest that sequential learning will not be adequate in tasks with complex spatial dependencies. Maintaining stable generalization is maintained by the Hybrid network that is reflected in the sharp rising trend of accuracy at the end epochs [12].

The precision-recall behavior analysis presented in Figure 2 and justified in Table 1 is also a good testimony to the increased ability to control errors in the Hybrid model. The Hybrid system was tested to have a better resilience to high precision without loss of recall when under test conditions with class imbalance. This property removes shared issues in the deep networks where the gains achieved in one measure adversely affect another. The fact that the CNN and LSTM models exhibit a minor range of variations shows that these models are slightly unstable, and therefore cannot be consistently reliable in terms of variations in the training data composition.

The relative overview that is offered in Table 2 goes further to reinforce the performance uniformity of the proposed Hybrid approach. The combination of the three different measures -AUC, F1 score, and log-loss- allows concluding that there is a coherent trend: Hybrid model shows its abilities in all aspects of statistics: it displays more accurate input behavior and gives more confidence-based predictions. Such strong results are symptomatic of scalable and stable architecture that can be used in applications where accuracy in classification, its sensitivity, and consistency of its probability are vital to the mission.

This paper has shown that the deep learning models, especially the hybrid LSTM-CNN architecture, have significant potential to forecast consumer decision-making in online marketing settings. Combination of sequential browsing data, sentiment, and demographic variables led to high performance with the model and provides valuable information on marketing optimization. The results point to the functional importance of deep learning integration into audience segmentation, personalized marketing, and prediction in planning a campaign.

Limitations of this study Practically, this research can be limited in the following way:

• Reliance on behavioral data of high volume and high quality that small companies might not be able to amass.
• Poor explainability of deep learning predictions, limiting its use in an environment where decisions need to be announced.
• The lower generalization of the model among the various industries with varied features of consumers.
• Legal and privacy implications that are involved in gathering consumer behavioral data.

Further directions need to concentrate to:

• Creating explainable artificial intelligence (XAI) solutions to transparent reasoning of prediction.
• Application of federated learning to provide the privacy of data during training of deep learning models over a distributed system.
• Incorporation of multimodal analytics that include a feature of audio sentiment, image engagement, and video interaction.
• Real-time adaptive learning: moving towards more responsiveness in the rapidly evolving digital markets.
• Learning about the transformer-based architecture that may help to model sequences better and understand the context.

Altogether, deep learning provides the prospective opportunities to the world of predicting consumer preferences and help marketers create more engaging, customized and efficient online designs.

1. K Bhavana Raj, Kamakshi Mehta, Someshwar Siddi, MK Sharma, Dilip K Sharma, Sunil Adhav, José LA Gonzáles, “Optimizing Financial Transactions and Processes Through the Power of Distributed Systems,” Meta Heuristic Algorithms for Advanced Distributed Systems, John Wiley & Sons, Inc, March 2024, https://doi.org/10.1002/9781394188093.ch17.
2. Abrokwah-Larbi, “The role of IoT and XAI convergence in the prediction, explanation, and decision of customer perceived value (CPV) in SMEs: a theoretical framework and research proposition perspective,” Discover Internet of Things, vol. 5, no. 1, Jan. 2025, doi: 10.1007/s43926-025-00092-x.
3. Imani, M. Joudaki, A. Beikmohammadi, and H. R. Arabnia, “Customer Churn Prediction: A systematic review of recent advances, trends, and challenges in machine learning and deep learning,” Machine Learning and Knowledge Extraction, vol. 7, no. 3, p. 105, Sep. 2025, doi: 10.3390/make7030105.
4. Razzaq and M. Shah, “Machine learning and deep learning paradigms: from techniques to practical applications and research frontiers,” Computers, vol. 14, no. 3, p. 93, Mar. 2025, doi: 10.3390/computers14030093.
5. Naz, H. U. Khan, A. Bukhari, B. Alshemaimri, A. Daud, and M. Ramzan, “Machine and deep learning for personality traits detection: a comprehensive survey and open research challenges,” Artificial Intelligence Review, vol. 58, no. 8, May 2025, doi: 1007/s10462-025-11245-3.
6. Alsharif, J. Wang, S. M. Isa, N. Z. M. Salleh, H. A. Dawas, and M. H. Alsharif, “The synergy of neuromarketing and artificial intelligence: A comprehensive literature review in the last decade,” Future Business Journal, vol. 11, no. 1, Jul. 2025, doi: 10.1186/s43093-025-00591-x.
7. O. Hassan and B. A. Hassan, “A comprehensive systematic review of machine learning in the retail industry: classifications, limitations, opportunities, and challenges,” Neural Computing and Applications, vol. 37, no. 4, pp. 2035–2070, Dec. 2024, doi: 10.1007/s00521-024-10869-w.
8. Riandhi, M. R. Arviansyah, and M. C. Sondari, “AI and consumer behavior: Trends, technologies, and future directions from a scopus-based systematic review,” Cogent Business & Management, vol. 12, no. 1, Aug. 2025, doi: 10.1080/23311975.2025.2544984.
9. N. Jawad and V. B. János, “‘A comprehensive review of AI-enhanced decision making: An empirical analysis for optimizing medication market business,’” Machine Learning With Applications, vol. 20, p. 100676, May 2025, doi: 10.1016/j.mlwa.2025.100676.
10. M.-D. Ablegue, M. Zhang, Q. Guo, and L. Rui, “Monitoring and shelf life extension of food products based on machine learning and deep learning: progress and potential applications,” Food Reviews International, pp. 1–26, Aug. 2025, doi: 10.1080/87559129.2025.2541856.
11. Oh, E. Ko, and M. Cho, “Fashion AI across the value chain: A comprehensive literature review and future agenda,” Journal of Global Scholars of Marketing Science, vol. 35, no. 4, pp. 516–538, Sep. 2025, doi: 10.1080/21639159.2025.2548816.
12. Braunsberger and E. Aschauer, “Corporate Failure Prediction: A literature review of Altman Z-Score and machine learning models within a technology adoption framework,” Journal of Risk and Financial Management, vol. 18, no. 8, p. 465, Aug. 2025, doi: 3390/jrfm18080465.
13. Someshwar Siddi, PM Sithar Selvam, "Intelligent Fuzzy Deep Learning Framework For Green Infrastructure Planning To Alleviate Urban Pollution And Climate Vulnerabilities", ICTACT Journal On Soft Computing, 16, 3, Oct 2025. doi: 10.21917/ijsc.2025.0553
14. E. Husseini, H. N. Noura, O. Salman, and K. Chahine, “Machine Learning in Smart Buildings: A review of methods, challenges, and future trends,” Applied Sciences, vol. 15, no. 14, p. 7682, Jul. 2025, doi: 10.3390/app15147682.
15. Y Raghunatha Reddy Someshwar Siddi, “Fuzzy Linear Problem Solution Using Lower Bound And Upper Bound Technique,” International Journal Of Mathematics And Physical Sciences Research ISSN 2348-5736 (Online) Vol. 5, 1, Pp: (1-8), Apr 2017