International Journal of Artificial Intelligence & Mathematical Sciences

A Machine Learning-Based Approach for Intelligent Game Recommendation System (I-GRS)

Mon, 20 Jul 2026 05:29:55 +0000

Hybrid game recommendation system would resolve the cold started problem on most recommendations engines, especially in digital gaming domain. General recommendation methods (Especially collaborative filtration) need users to be well interacted, which is not always the case. The hybrid model is the combination of collaborative filtration and content-based filtration to get the precision and recall on both new user case scenario as well as the experienced one. This model combines SVD matrix factorization with a gradient descent optimization which is well suited for sparse data in the user-item interaction matrix. At the same time, natural language processing (NLP) techniques analyze game plot descriptions so that the model can detect thematic similarities between games. This research proposed hybrid method on a data set of 787 games and found its results more valid to that of classical collaborative filtration–the mean precision is 0.015960 and the mean recall is 0.158242 and it also improves user recommendations for users that have interacted with very few items. Through 100 case studies with extensive records of game reviews, testing shows the model can operate within a sweet spot of recommendation accuracy and computational performance by performing best at approximately between 75–100 iterations. The combiner model is also responsible for more specificity and sensitivity than the best world records reported in benchmark datasets. Since game content and user preference change over time the system is suitable for online deployment as it interacts with users in real-time. User feedback and up-to-date content will drive its model, ensuring sustained interest as things to talk about change over time. To test with a large user base that includes the in-content groups and thus better constitute potentially cold-start users, as well as applying hybrid recommendation system techniques for use cases with other sparsity challenges. This research highlights the importance of combining collaborative filtration and content-based approaches for better recommendations, higher user satisfaction, and engagement on digital platforms.

Brain Stroke Prediction Using Artificial Intelligence and Machine Learning; A Comprehensive Risk Assessment Model

Saima Gul, Hafiz Syed Muhammad Kashif, Muhammad Yousuf Tufail — Mon, 20 Jul 2026 05:39:45 +0000

Brain stroke refers to a serious condition, happens when the blood flow to the brain is either blocked or affected owing to bleeding due to raptured vessel, in the brain. This lack of blood, thus lack of oxygen supply to brain cells harms brain tissue and leads to quick cell death. Each year, about fifteen million people around the world experienced the stroke. Of these, five million die, and another five million survive but face lasting disabilities. The mortality rate within 28 days after a stroke is about 28%. This study aims to develop a predictive model using artificial intelligence (AI) algorithms for assessing individual stroke risk. It uses a publicly available dataset of about 5,000 records from a website called Kaggle. The predictors include demographic and clinical variables including gender, hypertension, heart disease, smoking status etc. Seven machine learning classifiers, namely Decision Tree, Random Forest Classifier, Support Vector Machine SVM, Naïve Bayes, Logistic Regression, K-Nearest Neighbor KNN, and Gradient Boosting is implemented in Python. These classifiers are compared using different evaluation parameters. The result reveals that the Random Forest classifier consistently outperformed all other models in the evaluation parameters, achieving the highest predictive performance for stroke risk assessment.

Study the Prevalence of Upper Respiratory Tract Infections disease in high mountain area Hunza, Gilgit-Baltistan via Statistical Analysis

Mon, 20 Jul 2026 05:51:01 +0000

It is known that Upper respiratory tract infections (URTI) are among the most known causes of outpatient appointments globally. The current study investigates the prevalence of UTRI among population of Hunza District, Gilgit-Baltistan, may help to reduce the mortalities and comorbid health issues by taking prompt actions. The mountainous region Hunza is situated at high altitude, experienced cold weather conditions and have not better human health care facilitates. The continuous negligence may lead to the outbreak of this disease in whole Hunza District. A cross-sectional study and Exploratory Data Analysis (EDA) were applied to investigate the prevalence of UTRI and used monthly data from the year 1^st July- 2014 to 31-March -2016. The current study showed positive UTRI, 66.2% in age group 20 years, similarly 23.5% age group in 21-40 years old, further 6.9% in age group 41-60 years old whereas 3.20% in age groups of 61-80 years old. The overall gender-based prevalence of UTRI in all age group showed higher 53.1% in males as compared to 46.9% in females. URTI become burning health concerns all over the world, especially in less developing regions and must pay a high cost to the people of the society. The clinical study, significant results are convinced and suggested for further analysis of this disease on broad scale in the society.

Multi-Class Guava Classification Using Convolutional Neural Networks for Precision Agriculture

Mon, 20 Jul 2026 06:04:01 +0000

Guava (Psidium guajava L.) is an important fruit crop in Pakistan, especially in the province of Sindh where several varieties especially Thadharami, Local Sindhi and Riyali are grown and sold. While this is a common approach, the manual visual inspection approach is subjective, time-consuming, labor-intensive and prone to errors of classification, particularly when guava varieties and maturity stages have similar visual similarities. To overcome these, in this study, an automated image-based guava classification system has been proposed using Convolutional Neural Network (CNN). The proposed system differentiates between both guava variety and maturity stage (green, mature green and ripe). Images of guavas were arranged in nine classes of this data set according to its three local varieties and three maturity stages. The image was preprocessed to ensure uniformity of the input, minimize overfitting and improve the generalization of the model under variable imaging conditions. The CNN model was composed of several convolutional, pooling, and fully connected layers to extract the discriminative spatial, color, texture, and shape features in guava images automatically. The model is trained using categorical cross entropy loss and optimized with Adam optimizer. The experimental results presented here revealed the effectiveness of the proposed CNN for Guava classification by automatic classification with a validation accuracy of 92.5%. The misclassifications were higher for the green-stage classes as compared to the ripe guava classes, which was attributed to the large visual similarity between classes and high number of classes of green guavas. The result shows, CNN-based classification can be reliable, scalable and non-destructive solution for guava grading to ensure better quality control, decrease post-harvest losses, increase efficiency in supply chain of the guava industry in Pakistan.

Explainable Ai for Bone Marrow Cancer Diagnosis: from Bench to Bedside

Anuradha Reddy, Ochin Sharma, G G S Pradeep — Tue, 21 Jul 2026 05:24:37 +0000

Diagnosing bone marrow cancer continues to be a difficult process because of the large morphological variation of malignantly transformed cells and
insufficient interpretability of traditional deep learning approaches. We suggest a new efficient, interpretable, and clinically applicable artificial
intelligence (AI) framework for cancer diagnostics based on the combination of imaging and genomic data. Our framework uses INT8 quantization,
20% filter pruning, and dual-modality explainability, where SHAP is used to attribute genome sequencing features, and Grad-CAM++ is applied for
lesions localization based on MRI images. Performance evaluation is done by comparing the model to existing baselines, which include HA-UNet and
SE-ResHybrid models. According to our experiments, the suggested framework shows increased computational efficiency, decreasing the inference
time from 64 ms/image to 38 ms/image (by 41%), with a simultaneous improvement of segmentation (3.7%) and classification (4.5%) accuracy. In
addition, our model demonstrates high agreement with clinical specialists, achieving a Cohen's kappa value of 0.79, which proves its high clinical
relevancy. Localization accuracy also increases by 12-15%.