Explore research gaps
by sector
53 ML projects organized by sector, with research gap, dominant technique, and academic reference. Use it as a starting point for your thesis or research project.
Sector:
‹
›
| # | Sector | Project | Dominant Technique | Research Gap | Data Access | Review Reference |
|---|---|---|---|---|---|---|
| #3 | Health | GAN — Synthetic Mammograms Synthetic breast cancer image generation |
GANCNN | Limited clinical validation of synthetic images; risk of introducing artifacts that affect real diagnoses. | Medium |
Deep Learning in Medical Image Analysis (2025)
doi.org/10.1016/S1526-1492(25)00415-1
|
| #4 | Health | Leukemia — Blood Sample Classification Detection via microscopy images |
CNNTransfer Learning | Small datasets with limited ethnic diversity; poor generalization across different laboratories. | High |
Applying Deep Learning to Medical Imaging: A Review (2023)
doi.org/10.3390/app13189521
|
| #10 | Health | Psychological Risk Classifier App Triage to psychologist — AWS deployment |
Supervised MLNLP | Scarce validation in Latin American Spanish; lack of clinically labeled data for the region. | High |
Mental Health Prediction using ML: taxonomy & challenges (2022)
doi.org/10.1016/j.artmed.2022.102373
|
| #16 | Health | Diabetic Retinopathy — Image Detection Fundus retinal classification |
CNNResNet | Poor performance on low-quality images from rural clinics; limited explainability (XAI) for clinical staff. | Low |
Advances in AI for Medical Imaging (ScienceDirect, 2025)
doi.org/10.1016/j.procs.2025.09.457
|
| #24 | Health | Prostate Classification with Adversarial Attacks Model robustness under adversarial perturbations |
CNNAdversarial ML | Little research on robustness of medical models under attack; regulatory standards still undefined. | High |
XAI for Medical Imaging (Springer Cluster Computing, 2024)
doi.org/10.1007/s10586-025-05281-5
|
| #30 | Health | Hip Dysplasia Prediction — AWS Deployment Image classification + MongoDB + Lambda |
CNNCloud Deploy | Few studies with real production deployments; missing integration with existing hospital systems. | High |
Deep Learning in Medical Image Analysis (2025)
doi.org/10.1016/S1526-1492(25)00415-1
|
| #2 | Mining | Open-Pit Blast Fragmentation Explosive mesh design → fragment size prediction |
ML RegressionXGBoost | Scarce and proprietary data; limited IoT sensor integration for real-time dynamic adjustment. | High |
AI-Driven Predictive Maintenance in Mining (MDPI, 2025)
doi.org/10.3390/app15063337
|
| #5 | Mining | Grassland Degradation — Multimodal Classification Images + lab data + expert field observations |
MultimodalRF + CNN | Very few multimodal field-data models; scarce labeled Latin American datasets for this domain. | High |
ML Applications in Agriculture (Agronomy MDPI, 2023)
doi.org/10.3390/agronomy13122976
|
| #12 | Mining | Water Flow Prediction with LSTM Hydrological time series forecasting |
LSTMRNN | Poor generalization to untrained river basins; insufficient data quality in low-instrumentation regions. | Medium |
Transforming Mining Energy: ML Techniques (Frontiers, 2025)
doi.org/10.3389/fenrg.2025.1569716
|
| #13 | Mining | VAD — Energy Customer Segmentation Clustering by electricity consumption profile |
K-MeansDBSCAN | Limited advanced clustering in regulated tariffs; temporal behavioral variables rarely incorporated. | Medium |
Transforming Mining Energy: ML Techniques (Frontiers, 2025)
doi.org/10.3389/fenrg.2025.1569716
|
| #19 | Mining | SAG Mill Predictive Maintenance RUL / TTF / Failure window classification |
LSTMRFARIMA | Highly proprietary data; cross-mine model transfer still largely unexplored. | High |
AI-Driven Predictive Maintenance in Mining (MDPI, 2025)
doi.org/10.3390/app15063337
|
| #20 | Mining | Mining Operation Cost Prediction Daily cost forecasting — Huinchos mine |
ML RegressionGradient Boosting | Models with low interpretability for operators; scarce studies using real Latin American mining data. | High |
AI-Driven Predictive Maintenance in Mining (MDPI, 2025)
doi.org/10.3390/app15063337
|
| #25 | Mining | Energy Loss & Theft Detection Identification of non-technical loss zones |
ML ClassificationAnomaly Detection | Few studies with realistic synthetic data; explainable models for regulators and auditors still lacking. | Medium |
Transforming Mining Energy: ML Techniques (Frontiers, 2025)
doi.org/10.3389/fenrg.2025.1569716
|
| #27 | Mining | Rainfall Prediction with API Data Interactive meteorological data visualizations |
LSTMTime Series | Local models with low accuracy in mountainous regions; limited integration of satellite and reanalysis variables. | Low |
ML Applications in Agriculture (Agronomy MDPI, 2023)
doi.org/10.3390/agronomy13122976
|
| #35 | Mining | Mining Production Variable Optimization Identify key tasks that optimize production |
Feature ImportanceXGBoost | Non-standardized operational data across mines; little research on ML-based task sequence optimization. | High |
AI-Driven Predictive Maintenance in Mining (MDPI, 2025)
doi.org/10.3390/app15063337
|
| #41 | Mining | Electrical Failure Prediction from Historical Reports Predictive maintenance of electrical infrastructure |
ClassificationNLP on reports | Difficult labeling of past events; unstructured report text with non-standardized technical vocabulary. | High |
Predictive Maintenance: Bibliometric Analysis (JISEM, 2024)
doi.org/10.55267/iadt/09.jisem.2024
|
| #43 | Mining | Multimodal Geological Exploration Satellite imagery + geomagnetic + lab data |
Multimodal DLCNN + Tabular | Heterogeneous modality fusion largely unexplored; very limited public-domain geological datasets. | High |
AI-Driven Predictive Maintenance in Mining (MDPI, 2025)
doi.org/10.3390/app15063337
|
| #44 | Mining | Metaheuristics — ROP and MSE Prediction Rate of Penetration + Mechanical Specific Energy |
MetaheuristicsMultimodal Regression | Limited systematic comparison of metaheuristics in drilling; scarce open datasets for the sector. | High |
AI-Driven Predictive Maintenance in Mining (MDPI, 2025)
doi.org/10.3390/app15063337
|
| #34 | Mining | Mineralogical Arbitrator Quality Control Bias detection in mineral content assessments |
ClassificationFairness ML | Novel and understudied problem; no fairness frameworks applied to industrial inspection contexts. | High |
Unmasking Bias in AI: EHR-based models (ScienceDirect, 2024)
doi.org/10.1016/j.jbi.2024.104749
|
| #1 | Agriculture | Cement — Compressive Strength Prediction Regression on composition vs. strength |
ML RegressionSVR / ANN | Small homogeneous datasets; poor generalization to cements from different regions and formulations. | Low |
ML Applications in Agriculture (Agronomy MDPI, 2023)
doi.org/10.3390/agronomy13122976
|
| #2 | Agriculture | Rice — Variety Classification by Image Computer vision in agro-industry |
CNNTransfer Learning | Limited lighting and quality diversity in datasets; models not yet deployed on real production lines. | Low |
Applications of ML and DL in Agriculture (ScienceDirect, 2025)
doi.org/10.1016/j.atech.2025.100338
|
| #26 | Agriculture | Fish Farm Tank Variable Prediction (IoT) Aquaculture + real-time IoT API |
RegressionLSTMIoT | Very few ML studies in Latin American aquaculture; real-scale IoT-ML integration rarely validated. | High |
Applications of ML and DL in Agriculture (ScienceDirect, 2025)
doi.org/10.1016/j.atech.2025.100338
|
| #32 | Agriculture |
Grocery Price Prediction
Nutritional diet optimization from price forecasts
|
Regression Combinatorial Optimization | Combining price forecasting with nutritional optimization is novel; few open local price datasets available. | Medium |
ML in Agriculture: Impact on Supply Chain (Springer, 2025)
doi.org/10.1007/s44187-025-00419-1
|
| #33 | Agriculture | Cotton Pest Classification + BeeWare App Computer vision + native mobile deployment |
CNNYOLOMobile Deploy | Low-latency in-field pest detection apps; scarce Latin American pest image datasets. | Medium |
Applications of ML and DL in Agriculture (ScienceDirect, 2025)
doi.org/10.1016/j.atech.2025.100338
|
| #7 | Business | Business Data Analysis with Pandas Answering business questions with EDA |
EDAPandas | Introductory exercise; the real gap is automating actionable insights from EDA without manual intervention. | Low |
ML in Business and Finance: Literature Review (Springer, 2024)
doi.org/10.1186/s40854-024-00629-z
|
| #8 | Business | Sales Prediction, Segmentation & Anomaly Detection Full retail commercial analysis pipeline |
ClusteringRegressionAnomaly Detection | Integrating all three problems into a single unified pipeline is still largely unexplored in the literature. | Low |
ML in Business and Finance: Literature Review (Springer, 2024)
doi.org/10.1186/s40854-024-00629-z
|
| #15 | Business | Social Listening for Restaurant KPIs Reviews → business performance monitoring |
NLPSentiment AnalysisDashboard | Few studies connecting sentiment analysis to actionable KPIs in small food-service businesses. | Low |
NLP Sentiment Analysis: State-of-the-art Review (NLP Journal, 2024)
doi.org/10.1016/j.nlp.2024.100059
|
| #17 | Business | Markowitz with ML for Investment Portfolios Asset price prediction + portfolio optimization |
LSTMPortfolio Optimization | Limited validation on Latin American emerging markets; uncertainty not quantified in price forecasts. | Low |
Deep Learning in Finance: Survey (MDPI AI, 2024)
doi.org/10.3390/ai5040101
|
| #28 | Business | Employee Segmentation Unsupervised clustering of workforce profiles |
K-MeansPCAAdvanced Clustering | HR clustering in industrial contexts is understudied; performance-related features rarely included. | High |
ML in Business and Finance: Literature Review (Springer, 2024)
doi.org/10.1186/s40854-024-00629-z
|
| #29 | Business | Baseball Ticket Sales Prediction Point-of-sale estimation for ticket code release |
RegressionTime Series | Very specific business model (code release); little literature on sports ticket forecasting in LATAM. | High |
ML in Business and Finance: Literature Review (Springer, 2024)
doi.org/10.1186/s40854-024-00629-z
|
| #37 | Business | Hardware Store Sales Prediction Retail demand forecasting |
RegressionXGBoost | Local business data with heavy seasonality and outlier events; low availability of open sector data. | High |
ML in Business and Finance: Literature Review (Springer, 2024)
doi.org/10.1186/s40854-024-00629-z
|
| #45 | Business | US Bond Yield Prediction — Multiple Models Fixed-income return forecasting |
LSTMTransformersEnsemble | Impact of atypical macroeconomic events on time series; limited systematic architecture comparison. | Low |
Deep Learning in Finance: Survey (MDPI AI, 2024)
doi.org/10.3390/ai5040101
|
| #50 | Business | Corporate Innovation Policy Simulation DL on investment surveys + patent data |
DL on tabularScenario Simulation | ML-based policy simulation for business innovation is an emerging area with very little published literature. | High |
ML in Business and Finance: Literature Review (Springer, 2024)
doi.org/10.1186/s40854-024-00629-z
|
| #6 | Technology | Helmet Detection with YOLO Industrial safety — PPE detection |
YOLOObject Detection | High false positive rate in low-light conditions; poor generalization across different PPE types. | Low |
Deep Learning in Medical Image Analysis (ScienceDirect, 2025)
doi.org/10.1016/S1526-1492(25)00415-1
|
| #11 | Technology | CNN Models with VGG Base Image classification with VGG architecture |
VGGTransfer Learning | VGG is a mature architecture; the gap lies in combining it with attention mechanisms and parameter reduction. | Low |
Applying Deep Learning to Medical Imaging: A Review (MDPI, 2023)
doi.org/10.3390/app13189521
|
| #23 | Technology | Facial Recognition with Tkinter App Identification pipeline + desktop interface |
CNNFace Recognition | Racial and gender bias in recognition models; biometric data privacy and regulatory frameworks still evolving. | Low |
XAI for Medical Imaging (Springer Cluster Computing, 2024)
doi.org/10.1007/s10586-025-05281-5
|
| #38 | Technology | Cybersecurity — User Profiling & Attack Risk Cyberattack risk classification |
ML ClassificationAnomaly Detection | Lack of balanced real-attack datasets; models that poorly adapt to emerging threats (zero-day). | Medium |
AI Integration in Financial Services (Nature Humanities, 2025)
doi.org/10.1057/s41599-025-04850-8
|
| #40 | Technology | Violence Detection in Video Surveillance Security camera image classification |
CNNVideo Classification | High real-time latency; many false positives in crowded scenes or fast-motion environments. | Medium |
Applying Deep Learning to Medical Imaging: A Review (MDPI, 2023)
doi.org/10.3390/app13189521
|
| #44b | Technology | PPE Detection — Full IoT Project Industrial safety + complete IoT architecture |
YOLOIoT Edge | Edge deployment on constrained hardware; limited work on adapting YOLO to embedded IoT devices. | Medium |
AI-Driven Predictive Maintenance in Mining (MDPI, 2025)
doi.org/10.3390/app15063337
|
| #48 | Technology | PID Optimization with Metaheuristics Automatic PID controller tuning |
MetaheuristicsGA / PSO | Insufficient systematic comparison of metaheuristic algorithms for PID; limited validation on real systems. | High |
AI-Driven Predictive Maintenance in Mining (MDPI, 2025)
doi.org/10.3390/app15063337
|
| #9 | Gov / Social | Social Listening — COVID-19 Vaccine Opinion (Peru) Tweets + sentiment analysis in public health |
NLPSentiment AnalysisTwitter API | Limited coverage of Latin American Spanish; Twitter data now heavily restricted and costly post-X/Elon. | High |
NLP Sentiment Analysis: State-of-the-art Review (NLP Journal, 2024)
doi.org/10.1016/j.nlp.2024.100059
|
| #14 | Gov / Social | Sentiment Stream with ElasticSearch Dashboard + continuous tweet stream |
NLPStreamingElasticSearch | Real-time NLP architectures still inaccessible for small institutions and local governments. | Medium |
Challenges in Deep Learning for Sentiment Analysis (Springer AI Review, 2024)
doi.org/10.1007/s10462-023-10651-9
|
| #39 | Gov / Social | Domestic Violence — Police Report Classification NLP on text + multiclass classification |
NLPML Classification | Informal police language; high ethical sensitivity; very few open Spanish-language labeled datasets. | High |
NLP Sentiment Analysis: State-of-the-art Review (NLP Journal, 2024)
doi.org/10.1016/j.nlp.2024.100059
|
| #42 | Gov / Social | Social Listening: NLP + Structural Equation Models SEM + NLP for marketing decisions |
NLPSEMClassification | NLP + SEM combination is methodologically complex; rarely explored in Latin American marketing contexts. | Medium |
Challenges in Deep Learning for Sentiment Analysis (Springer AI Review, 2024)
doi.org/10.1007/s10462-023-10651-9
|
| #43b | Gov / Social | Social Listening: YouTube & Trending News Topic classification in YouTube comments |
NLPTopic ModelingBERT | Real-time trend detection on video platforms still has high NLP latency and limited coverage. | Low |
NLP Sentiment Analysis: State-of-the-art Review (NLP Journal, 2024)
doi.org/10.1016/j.nlp.2024.100059
|
| #49 | Gov / Social | Social Listening: Tourism (TripAdvisor) Tourism review analysis in Spanish |
NLPSentiment Analysis | Few NLP studies on tourism in emerging Latin American destinations in Spanish. | Low |
NLP Sentiment Analysis: State-of-the-art Review (NLP Journal, 2024)
doi.org/10.1016/j.nlp.2024.100059
|
| #51 | Gov / Social | Qualitative Analysis for Public Policy Social sciences + AI-assisted policy generation |
NLPAI-assisted qualitative | AI for qualitative social science analysis is still emerging; risk of losing interpretive nuance. | High |
Challenges in Deep Learning for Sentiment Analysis (Springer AI Review, 2024)
doi.org/10.1007/s10462-023-10651-9
|
| #3b | Gov / Social | OCR on National ID Cards (Pytesseract) Character detection in identity documents |
OCRPytesseractCV | Low accuracy on damaged or typographically varied documents; poor generalization across countries and formats. | High |
Applying Deep Learning to Medical Imaging: A Review (MDPI, 2023)
doi.org/10.3390/app13189521
|
| #21 | Industry | Roller Wear — Manufacturing Component wear prediction in industrial machinery |
ML RegressionFeature Engineering | Proprietary data hard to share; need for realistic synthetic datasets to generalize models across facilities. | High |
Predictive Maintenance: Bibliometric Analysis (JISEM, 2024)
doi.org/10.55267/iadt/09.jisem.2024
|
| #22 | Industry | Request Classification with NLP (Customer Service) Free-text notes → department routing |
NLPText ClassificationBERT | Industrial technical vocabulary requires costly fine-tuning; scarce labeled data in domain-specific contexts. | High |
NLP Sentiment Analysis: State-of-the-art Review (NLP Journal, 2024)
doi.org/10.1016/j.nlp.2024.100059
|
| #31 | Industry | Bridge Damage Classification — Images Visual inspection of civil infrastructure |
CNNObject Detection | Datasets lack weather condition and damage type variety; no integration with asset management systems. | Medium |
Deep Learning in Medical Image Analysis (ScienceDirect, 2025)
doi.org/10.1016/S1526-1492(25)00415-1
|
| #18 | Entertainment / AI | Space Invaders with DQN Deep reinforcement learning in video games |
DQNReinforcement Learning | Transferring game-learned policies to real-world environments; sample efficiency still very low. | Low |
Deep Learning in Finance: Survey (MDPI AI, 2024)
doi.org/10.3390/ai5040101
|
| #36 | Academic Research | Individual Treatment Effects (ITE) with ML Causal inference + ML for public policy |
Causal MLITE estimation | ITE estimation in small populations; limited validation in Latin American public policy contexts. | High |
ML in Business and Finance: Literature Review (Springer, 2024)
doi.org/10.1186/s40854-024-00629-z
|